1 /* SPDX-License-Identifier: LGPL-2.1+ */
9 #include <sys/statvfs.h>
15 #include "alloc-util.h"
16 #include "btrfs-util.h"
17 #include "chattr-util.h"
20 #include "format-util.h"
22 #include "journal-authenticate.h"
23 #include "journal-def.h"
24 #include "journal-file.h"
26 #include "memory-util.h"
27 #include "path-util.h"
28 #include "random-util.h"
30 #include "sort-util.h"
31 #include "stat-util.h"
32 #include "string-util.h"
34 #include "xattr-util.h"
36 #define DEFAULT_DATA_HASH_TABLE_SIZE (2047ULL*sizeof(HashItem))
37 #define DEFAULT_FIELD_HASH_TABLE_SIZE (333ULL*sizeof(HashItem))
39 #define DEFAULT_COMPRESS_THRESHOLD (512ULL)
40 #define MIN_COMPRESS_THRESHOLD (8ULL)
42 /* This is the minimum journal file size */
43 #define JOURNAL_FILE_SIZE_MIN (512 * 1024ULL) /* 512 KiB */
45 /* These are the lower and upper bounds if we deduce the max_use value
46 * from the file system size */
47 #define MAX_USE_LOWER (1 * 1024 * 1024ULL) /* 1 MiB */
48 #define MAX_USE_UPPER (4 * 1024 * 1024 * 1024ULL) /* 4 GiB */
50 /* Those are the lower and upper bounds for the minimal use limit,
51 * i.e. how much we'll use even if keep_free suggests otherwise. */
52 #define MIN_USE_LOW (1 * 1024 * 1024ULL) /* 1 MiB */
53 #define MIN_USE_HIGH (16 * 1024 * 1024ULL) /* 16 MiB */
55 /* This is the upper bound if we deduce max_size from max_use */
56 #define MAX_SIZE_UPPER (128 * 1024 * 1024ULL) /* 128 MiB */
58 /* This is the upper bound if we deduce the keep_free value from the
60 #define KEEP_FREE_UPPER (4 * 1024 * 1024 * 1024ULL) /* 4 GiB */
62 /* This is the keep_free value when we can't determine the system
64 #define DEFAULT_KEEP_FREE (1024 * 1024ULL) /* 1 MB */
66 /* This is the default maximum number of journal files to keep around. */
67 #define DEFAULT_N_MAX_FILES 100
69 /* n_data was the first entry we added after the initial file format design */
70 #define HEADER_SIZE_MIN ALIGN64(offsetof(Header, n_data))
72 /* How many entries to keep in the entry array chain cache at max */
73 #define CHAIN_CACHE_MAX 20
75 /* How much to increase the journal file size at once each time we allocate something new. */
76 #define FILE_SIZE_INCREASE (8 * 1024 * 1024ULL) /* 8MB */
78 /* Reread fstat() of the file for detecting deletions at least this often */
79 #define LAST_STAT_REFRESH_USEC (5*USEC_PER_SEC)
81 /* The mmap context to use for the header we pick as one above the last defined typed */
82 #define CONTEXT_HEADER _OBJECT_TYPE_MAX
85 # pragma GCC diagnostic ignored "-Waddress-of-packed-member"
88 /* This may be called from a separate thread to prevent blocking the caller for the duration of fsync().
89 * As a result we use atomic operations on f->offline_state for inter-thread communications with
90 * journal_file_set_offline() and journal_file_set_online(). */
91 static void journal_file_set_offline_internal(JournalFile
*f
) {
97 switch (f
->offline_state
) {
99 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_CANCEL
, OFFLINE_DONE
))
103 case OFFLINE_AGAIN_FROM_SYNCING
:
104 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_AGAIN_FROM_SYNCING
, OFFLINE_SYNCING
))
108 case OFFLINE_AGAIN_FROM_OFFLINING
:
109 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_AGAIN_FROM_OFFLINING
, OFFLINE_SYNCING
))
113 case OFFLINE_SYNCING
:
116 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_SYNCING
, OFFLINE_OFFLINING
))
119 f
->header
->state
= f
->archive
? STATE_ARCHIVED
: STATE_OFFLINE
;
123 case OFFLINE_OFFLINING
:
124 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_OFFLINING
, OFFLINE_DONE
))
131 log_debug("OFFLINE_JOINED unexpected offline state for journal_file_set_offline_internal()");
137 static void * journal_file_set_offline_thread(void *arg
) {
138 JournalFile
*f
= arg
;
140 (void) pthread_setname_np(pthread_self(), "journal-offline");
142 journal_file_set_offline_internal(f
);
147 static int journal_file_set_offline_thread_join(JournalFile
*f
) {
152 if (f
->offline_state
== OFFLINE_JOINED
)
155 r
= pthread_join(f
->offline_thread
, NULL
);
159 f
->offline_state
= OFFLINE_JOINED
;
161 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
167 /* Trigger a restart if the offline thread is mid-flight in a restartable state. */
168 static bool journal_file_set_offline_try_restart(JournalFile
*f
) {
170 switch (f
->offline_state
) {
171 case OFFLINE_AGAIN_FROM_SYNCING
:
172 case OFFLINE_AGAIN_FROM_OFFLINING
:
176 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_CANCEL
, OFFLINE_AGAIN_FROM_SYNCING
))
180 case OFFLINE_SYNCING
:
181 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_SYNCING
, OFFLINE_AGAIN_FROM_SYNCING
))
185 case OFFLINE_OFFLINING
:
186 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_OFFLINING
, OFFLINE_AGAIN_FROM_OFFLINING
))
196 /* Sets a journal offline.
198 * If wait is false then an offline is dispatched in a separate thread for a
199 * subsequent journal_file_set_offline() or journal_file_set_online() of the
200 * same journal to synchronize with.
202 * If wait is true, then either an existing offline thread will be restarted
203 * and joined, or if none exists the offline is simply performed in this
204 * context without involving another thread.
206 int journal_file_set_offline(JournalFile
*f
, bool wait
) {
215 if (f
->fd
< 0 || !f
->header
)
218 /* An offlining journal is implicitly online and may modify f->header->state,
219 * we must also join any potentially lingering offline thread when not online. */
220 if (!journal_file_is_offlining(f
) && f
->header
->state
!= STATE_ONLINE
)
221 return journal_file_set_offline_thread_join(f
);
223 /* Restart an in-flight offline thread and wait if needed, or join a lingering done one. */
224 restarted
= journal_file_set_offline_try_restart(f
);
225 if ((restarted
&& wait
) || !restarted
) {
226 r
= journal_file_set_offline_thread_join(f
);
234 /* Initiate a new offline. */
235 f
->offline_state
= OFFLINE_SYNCING
;
237 if (wait
) /* Without using a thread if waiting. */
238 journal_file_set_offline_internal(f
);
240 sigset_t ss
, saved_ss
;
243 assert_se(sigfillset(&ss
) >= 0);
244 /* Don't block SIGBUS since the offlining thread accesses a memory mapped file.
245 * Asynchronous SIGBUS signals can safely be handled by either thread. */
246 assert_se(sigdelset(&ss
, SIGBUS
) >= 0);
248 r
= pthread_sigmask(SIG_BLOCK
, &ss
, &saved_ss
);
252 r
= pthread_create(&f
->offline_thread
, NULL
, journal_file_set_offline_thread
, f
);
254 k
= pthread_sigmask(SIG_SETMASK
, &saved_ss
, NULL
);
256 f
->offline_state
= OFFLINE_JOINED
;
266 static int journal_file_set_online(JournalFile
*f
) {
274 if (f
->fd
< 0 || !f
->header
)
278 switch (f
->offline_state
) {
280 /* No offline thread, no need to wait. */
284 case OFFLINE_SYNCING
:
285 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_SYNCING
, OFFLINE_CANCEL
))
287 /* Canceled syncing prior to offlining, no need to wait. */
291 case OFFLINE_AGAIN_FROM_SYNCING
:
292 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_AGAIN_FROM_SYNCING
, OFFLINE_CANCEL
))
294 /* Canceled restart from syncing, no need to wait. */
298 case OFFLINE_AGAIN_FROM_OFFLINING
:
299 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_AGAIN_FROM_OFFLINING
, OFFLINE_CANCEL
))
301 /* Canceled restart from offlining, must wait for offlining to complete however. */
306 r
= journal_file_set_offline_thread_join(f
);
316 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
319 switch (f
->header
->state
) {
324 f
->header
->state
= STATE_ONLINE
;
333 bool journal_file_is_offlining(JournalFile
*f
) {
336 __sync_synchronize();
338 if (IN_SET(f
->offline_state
, OFFLINE_DONE
, OFFLINE_JOINED
))
344 JournalFile
* journal_file_close(JournalFile
*f
) {
349 /* Write the final tag */
350 if (f
->seal
&& f
->writable
) {
353 r
= journal_file_append_tag(f
);
355 log_error_errno(r
, "Failed to append tag when closing journal: %m");
359 if (f
->post_change_timer
) {
360 if (sd_event_source_get_enabled(f
->post_change_timer
, NULL
) > 0)
361 journal_file_post_change(f
);
363 sd_event_source_disable_unref(f
->post_change_timer
);
366 journal_file_set_offline(f
, true);
368 if (f
->mmap
&& f
->cache_fd
)
369 mmap_cache_free_fd(f
->mmap
, f
->cache_fd
);
371 if (f
->fd
>= 0 && f
->defrag_on_close
) {
373 /* Be friendly to btrfs: turn COW back on again now,
374 * and defragment the file. We won't write to the file
375 * ever again, hence remove all fragmentation, and
376 * reenable all the good bits COW usually provides
377 * (such as data checksumming). */
379 (void) chattr_fd(f
->fd
, 0, FS_NOCOW_FL
, NULL
);
380 (void) btrfs_defrag_fd(f
->fd
);
387 mmap_cache_unref(f
->mmap
);
389 ordered_hashmap_free_free(f
->chain_cache
);
391 #if HAVE_XZ || HAVE_LZ4
392 free(f
->compress_buffer
);
397 munmap(f
->fss_file
, PAGE_ALIGN(f
->fss_file_size
));
399 free(f
->fsprg_state
);
404 gcry_md_close(f
->hmac
);
410 static int journal_file_init_header(JournalFile
*f
, JournalFile
*template) {
417 memcpy(h
.signature
, HEADER_SIGNATURE
, 8);
418 h
.header_size
= htole64(ALIGN64(sizeof(h
)));
420 h
.incompatible_flags
|= htole32(
421 f
->compress_xz
* HEADER_INCOMPATIBLE_COMPRESSED_XZ
|
422 f
->compress_lz4
* HEADER_INCOMPATIBLE_COMPRESSED_LZ4
);
424 h
.compatible_flags
= htole32(
425 f
->seal
* HEADER_COMPATIBLE_SEALED
);
427 r
= sd_id128_randomize(&h
.file_id
);
432 h
.seqnum_id
= template->header
->seqnum_id
;
433 h
.tail_entry_seqnum
= template->header
->tail_entry_seqnum
;
435 h
.seqnum_id
= h
.file_id
;
437 k
= pwrite(f
->fd
, &h
, sizeof(h
), 0);
447 static int journal_file_refresh_header(JournalFile
*f
) {
453 r
= sd_id128_get_machine(&f
->header
->machine_id
);
454 if (IN_SET(r
, -ENOENT
, -ENOMEDIUM
))
455 /* We don't have a machine-id, let's continue without */
456 zero(f
->header
->machine_id
);
460 r
= sd_id128_get_boot(&f
->header
->boot_id
);
464 r
= journal_file_set_online(f
);
466 /* Sync the online state to disk */
469 /* We likely just created a new file, also sync the directory this file is located in. */
470 (void) fsync_directory_of_file(f
->fd
);
475 static bool warn_wrong_flags(const JournalFile
*f
, bool compatible
) {
476 const uint32_t any
= compatible
? HEADER_COMPATIBLE_ANY
: HEADER_INCOMPATIBLE_ANY
,
477 supported
= compatible
? HEADER_COMPATIBLE_SUPPORTED
: HEADER_INCOMPATIBLE_SUPPORTED
;
478 const char *type
= compatible
? "compatible" : "incompatible";
481 flags
= le32toh(compatible
? f
->header
->compatible_flags
: f
->header
->incompatible_flags
);
483 if (flags
& ~supported
) {
485 log_debug("Journal file %s has unknown %s flags 0x%"PRIx32
,
486 f
->path
, type
, flags
& ~any
);
487 flags
= (flags
& any
) & ~supported
;
491 _cleanup_free_
char *t
= NULL
;
493 if (compatible
&& (flags
& HEADER_COMPATIBLE_SEALED
))
494 strv
[n
++] = "sealed";
495 if (!compatible
&& (flags
& HEADER_INCOMPATIBLE_COMPRESSED_XZ
))
496 strv
[n
++] = "xz-compressed";
497 if (!compatible
&& (flags
& HEADER_INCOMPATIBLE_COMPRESSED_LZ4
))
498 strv
[n
++] = "lz4-compressed";
500 assert(n
< ELEMENTSOF(strv
));
502 t
= strv_join((char**) strv
, ", ");
503 log_debug("Journal file %s uses %s %s %s disabled at compilation time.",
504 f
->path
, type
, n
> 1 ? "flags" : "flag", strnull(t
));
512 static int journal_file_verify_header(JournalFile
*f
) {
513 uint64_t arena_size
, header_size
;
518 if (memcmp(f
->header
->signature
, HEADER_SIGNATURE
, 8))
521 /* In both read and write mode we refuse to open files with incompatible
522 * flags we don't know. */
523 if (warn_wrong_flags(f
, false))
524 return -EPROTONOSUPPORT
;
526 /* When open for writing we refuse to open files with compatible flags, too. */
527 if (f
->writable
&& warn_wrong_flags(f
, true))
528 return -EPROTONOSUPPORT
;
530 if (f
->header
->state
>= _STATE_MAX
)
533 header_size
= le64toh(READ_NOW(f
->header
->header_size
));
535 /* The first addition was n_data, so check that we are at least this large */
536 if (header_size
< HEADER_SIZE_MIN
)
539 if (JOURNAL_HEADER_SEALED(f
->header
) && !JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
542 arena_size
= le64toh(READ_NOW(f
->header
->arena_size
));
544 if (UINT64_MAX
- header_size
< arena_size
|| header_size
+ arena_size
> (uint64_t) f
->last_stat
.st_size
)
547 if (le64toh(f
->header
->tail_object_offset
) > header_size
+ arena_size
)
550 if (!VALID64(le64toh(f
->header
->data_hash_table_offset
)) ||
551 !VALID64(le64toh(f
->header
->field_hash_table_offset
)) ||
552 !VALID64(le64toh(f
->header
->tail_object_offset
)) ||
553 !VALID64(le64toh(f
->header
->entry_array_offset
)))
557 sd_id128_t machine_id
;
561 r
= sd_id128_get_machine(&machine_id
);
565 if (!sd_id128_equal(machine_id
, f
->header
->machine_id
))
568 state
= f
->header
->state
;
570 if (state
== STATE_ARCHIVED
)
571 return -ESHUTDOWN
; /* Already archived */
572 else if (state
== STATE_ONLINE
)
573 return log_debug_errno(SYNTHETIC_ERRNO(EBUSY
),
574 "Journal file %s is already online. Assuming unclean closing.",
576 else if (state
!= STATE_OFFLINE
)
577 return log_debug_errno(SYNTHETIC_ERRNO(EBUSY
),
578 "Journal file %s has unknown state %i.",
581 if (f
->header
->field_hash_table_size
== 0 || f
->header
->data_hash_table_size
== 0)
584 /* Don't permit appending to files from the future. Because otherwise the realtime timestamps wouldn't
585 * be strictly ordered in the entries in the file anymore, and we can't have that since it breaks
587 if (le64toh(f
->header
->tail_entry_realtime
) > now(CLOCK_REALTIME
))
588 return log_debug_errno(SYNTHETIC_ERRNO(ETXTBSY
),
589 "Journal file %s is from the future, refusing to append new data to it that'd be older.",
593 f
->compress_xz
= JOURNAL_HEADER_COMPRESSED_XZ(f
->header
);
594 f
->compress_lz4
= JOURNAL_HEADER_COMPRESSED_LZ4(f
->header
);
596 f
->seal
= JOURNAL_HEADER_SEALED(f
->header
);
601 int journal_file_fstat(JournalFile
*f
) {
607 if (fstat(f
->fd
, &f
->last_stat
) < 0)
610 f
->last_stat_usec
= now(CLOCK_MONOTONIC
);
612 /* Refuse dealing with with files that aren't regular */
613 r
= stat_verify_regular(&f
->last_stat
);
617 /* Refuse appending to files that are already deleted */
618 if (f
->last_stat
.st_nlink
<= 0)
624 static int journal_file_allocate(JournalFile
*f
, uint64_t offset
, uint64_t size
) {
625 uint64_t old_size
, new_size
, old_header_size
, old_arena_size
;
631 /* We assume that this file is not sparse, and we know that for sure, since we always call
632 * posix_fallocate() ourselves */
634 if (size
> PAGE_ALIGN_DOWN(UINT64_MAX
) - offset
)
637 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
640 old_header_size
= le64toh(READ_NOW(f
->header
->header_size
));
641 old_arena_size
= le64toh(READ_NOW(f
->header
->arena_size
));
642 if (old_arena_size
> PAGE_ALIGN_DOWN(UINT64_MAX
) - old_header_size
)
645 old_size
= old_header_size
+ old_arena_size
;
647 new_size
= MAX(PAGE_ALIGN(offset
+ size
), old_header_size
);
649 if (new_size
<= old_size
) {
651 /* We already pre-allocated enough space, but before
652 * we write to it, let's check with fstat() if the
653 * file got deleted, in order make sure we don't throw
654 * away the data immediately. Don't check fstat() for
655 * all writes though, but only once ever 10s. */
657 if (f
->last_stat_usec
+ LAST_STAT_REFRESH_USEC
> now(CLOCK_MONOTONIC
))
660 return journal_file_fstat(f
);
663 /* Allocate more space. */
665 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
668 if (new_size
> f
->metrics
.min_size
&& f
->metrics
.keep_free
> 0) {
671 if (fstatvfs(f
->fd
, &svfs
) >= 0) {
674 available
= LESS_BY((uint64_t) svfs
.f_bfree
* (uint64_t) svfs
.f_bsize
, f
->metrics
.keep_free
);
676 if (new_size
- old_size
> available
)
681 /* Increase by larger blocks at once */
682 new_size
= DIV_ROUND_UP(new_size
, FILE_SIZE_INCREASE
) * FILE_SIZE_INCREASE
;
683 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
684 new_size
= f
->metrics
.max_size
;
686 /* Note that the glibc fallocate() fallback is very
687 inefficient, hence we try to minimize the allocation area
689 r
= posix_fallocate(f
->fd
, old_size
, new_size
- old_size
);
693 f
->header
->arena_size
= htole64(new_size
- old_header_size
);
695 return journal_file_fstat(f
);
698 static unsigned type_to_context(ObjectType type
) {
699 /* One context for each type, plus one catch-all for the rest */
700 assert_cc(_OBJECT_TYPE_MAX
<= MMAP_CACHE_MAX_CONTEXTS
);
701 assert_cc(CONTEXT_HEADER
< MMAP_CACHE_MAX_CONTEXTS
);
702 return type
> OBJECT_UNUSED
&& type
< _OBJECT_TYPE_MAX
? type
: 0;
705 static int journal_file_move_to(
722 if (size
> UINT64_MAX
- offset
)
725 /* Avoid SIGBUS on invalid accesses */
726 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
) {
727 /* Hmm, out of range? Let's refresh the fstat() data
728 * first, before we trust that check. */
730 r
= journal_file_fstat(f
);
734 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
)
735 return -EADDRNOTAVAIL
;
738 return mmap_cache_get(f
->mmap
, f
->cache_fd
, f
->prot
, type_to_context(type
), keep_always
, offset
, size
, &f
->last_stat
, ret
, ret_size
);
741 static uint64_t minimum_header_size(Object
*o
) {
743 static const uint64_t table
[] = {
744 [OBJECT_DATA
] = sizeof(DataObject
),
745 [OBJECT_FIELD
] = sizeof(FieldObject
),
746 [OBJECT_ENTRY
] = sizeof(EntryObject
),
747 [OBJECT_DATA_HASH_TABLE
] = sizeof(HashTableObject
),
748 [OBJECT_FIELD_HASH_TABLE
] = sizeof(HashTableObject
),
749 [OBJECT_ENTRY_ARRAY
] = sizeof(EntryArrayObject
),
750 [OBJECT_TAG
] = sizeof(TagObject
),
753 if (o
->object
.type
>= ELEMENTSOF(table
) || table
[o
->object
.type
] <= 0)
754 return sizeof(ObjectHeader
);
756 return table
[o
->object
.type
];
759 /* Lightweight object checks. We want this to be fast, so that we won't
760 * slowdown every journal_file_move_to_object() call too much. */
761 static int journal_file_check_object(JournalFile
*f
, uint64_t offset
, Object
*o
) {
765 switch (o
->object
.type
) {
768 if ((le64toh(o
->data
.entry_offset
) == 0) ^ (le64toh(o
->data
.n_entries
) == 0))
769 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
770 "Bad n_entries: %" PRIu64
": %" PRIu64
,
771 le64toh(o
->data
.n_entries
),
774 if (le64toh(o
->object
.size
) <= offsetof(DataObject
, payload
))
775 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
776 "Bad object size (<= %zu): %" PRIu64
": %" PRIu64
,
777 offsetof(DataObject
, payload
),
778 le64toh(o
->object
.size
),
781 if (!VALID64(le64toh(o
->data
.next_hash_offset
)) ||
782 !VALID64(le64toh(o
->data
.next_field_offset
)) ||
783 !VALID64(le64toh(o
->data
.entry_offset
)) ||
784 !VALID64(le64toh(o
->data
.entry_array_offset
)))
785 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
786 "Invalid offset, next_hash_offset=" OFSfmt
", next_field_offset=" OFSfmt
", entry_offset=" OFSfmt
", entry_array_offset=" OFSfmt
": %" PRIu64
,
787 le64toh(o
->data
.next_hash_offset
),
788 le64toh(o
->data
.next_field_offset
),
789 le64toh(o
->data
.entry_offset
),
790 le64toh(o
->data
.entry_array_offset
),
796 if (le64toh(o
->object
.size
) <= offsetof(FieldObject
, payload
))
797 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
798 "Bad field size (<= %zu): %" PRIu64
": %" PRIu64
,
799 offsetof(FieldObject
, payload
),
800 le64toh(o
->object
.size
),
803 if (!VALID64(le64toh(o
->field
.next_hash_offset
)) ||
804 !VALID64(le64toh(o
->field
.head_data_offset
)))
805 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
806 "Invalid offset, next_hash_offset=" OFSfmt
", head_data_offset=" OFSfmt
": %" PRIu64
,
807 le64toh(o
->field
.next_hash_offset
),
808 le64toh(o
->field
.head_data_offset
),
815 sz
= le64toh(READ_NOW(o
->object
.size
));
816 if (sz
< offsetof(EntryObject
, items
) ||
817 (sz
- offsetof(EntryObject
, items
)) % sizeof(EntryItem
) != 0)
818 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
819 "Bad entry size (<= %zu): %" PRIu64
": %" PRIu64
,
820 offsetof(EntryObject
, items
),
824 if ((sz
- offsetof(EntryObject
, items
)) / sizeof(EntryItem
) <= 0)
825 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
826 "Invalid number items in entry: %" PRIu64
": %" PRIu64
,
827 (sz
- offsetof(EntryObject
, items
)) / sizeof(EntryItem
),
830 if (le64toh(o
->entry
.seqnum
) <= 0)
831 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
832 "Invalid entry seqnum: %" PRIx64
": %" PRIu64
,
833 le64toh(o
->entry
.seqnum
),
836 if (!VALID_REALTIME(le64toh(o
->entry
.realtime
)))
837 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
838 "Invalid entry realtime timestamp: %" PRIu64
": %" PRIu64
,
839 le64toh(o
->entry
.realtime
),
842 if (!VALID_MONOTONIC(le64toh(o
->entry
.monotonic
)))
843 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
844 "Invalid entry monotonic timestamp: %" PRIu64
": %" PRIu64
,
845 le64toh(o
->entry
.monotonic
),
851 case OBJECT_DATA_HASH_TABLE
:
852 case OBJECT_FIELD_HASH_TABLE
: {
855 sz
= le64toh(READ_NOW(o
->object
.size
));
856 if (sz
< offsetof(HashTableObject
, items
) ||
857 (sz
- offsetof(HashTableObject
, items
)) % sizeof(HashItem
) != 0 ||
858 (sz
- offsetof(HashTableObject
, items
)) / sizeof(HashItem
) <= 0)
859 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
860 "Invalid %s hash table size: %" PRIu64
": %" PRIu64
,
861 o
->object
.type
== OBJECT_DATA_HASH_TABLE
? "data" : "field",
868 case OBJECT_ENTRY_ARRAY
: {
871 sz
= le64toh(READ_NOW(o
->object
.size
));
872 if (sz
< offsetof(EntryArrayObject
, items
) ||
873 (sz
- offsetof(EntryArrayObject
, items
)) % sizeof(le64_t
) != 0 ||
874 (sz
- offsetof(EntryArrayObject
, items
)) / sizeof(le64_t
) <= 0)
875 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
876 "Invalid object entry array size: %" PRIu64
": %" PRIu64
,
880 if (!VALID64(le64toh(o
->entry_array
.next_entry_array_offset
)))
881 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
882 "Invalid object entry array next_entry_array_offset: " OFSfmt
": %" PRIu64
,
883 le64toh(o
->entry_array
.next_entry_array_offset
),
890 if (le64toh(o
->object
.size
) != sizeof(TagObject
))
891 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
892 "Invalid object tag size: %" PRIu64
": %" PRIu64
,
893 le64toh(o
->object
.size
),
896 if (!VALID_EPOCH(le64toh(o
->tag
.epoch
)))
897 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
898 "Invalid object tag epoch: %" PRIu64
": %" PRIu64
,
899 le64toh(o
->tag
.epoch
), offset
);
907 int journal_file_move_to_object(JournalFile
*f
, ObjectType type
, uint64_t offset
, Object
**ret
) {
917 /* Objects may only be located at multiple of 64 bit */
918 if (!VALID64(offset
))
919 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
920 "Attempt to move to object at non-64bit boundary: %" PRIu64
,
923 /* Object may not be located in the file header */
924 if (offset
< le64toh(f
->header
->header_size
))
925 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
926 "Attempt to move to object located in file header: %" PRIu64
,
929 r
= journal_file_move_to(f
, type
, false, offset
, sizeof(ObjectHeader
), &t
, &tsize
);
934 s
= le64toh(READ_NOW(o
->object
.size
));
937 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
938 "Attempt to move to uninitialized object: %" PRIu64
,
940 if (s
< sizeof(ObjectHeader
))
941 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
942 "Attempt to move to overly short object: %" PRIu64
,
945 if (o
->object
.type
<= OBJECT_UNUSED
)
946 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
947 "Attempt to move to object with invalid type: %" PRIu64
,
950 if (s
< minimum_header_size(o
))
951 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
952 "Attempt to move to truncated object: %" PRIu64
,
955 if (type
> OBJECT_UNUSED
&& o
->object
.type
!= type
)
956 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
957 "Attempt to move to object of unexpected type: %" PRIu64
,
961 r
= journal_file_move_to(f
, type
, false, offset
, s
, &t
, NULL
);
968 r
= journal_file_check_object(f
, offset
, o
);
976 static uint64_t journal_file_entry_seqnum(JournalFile
*f
, uint64_t *seqnum
) {
982 r
= le64toh(f
->header
->tail_entry_seqnum
) + 1;
985 /* If an external seqnum counter was passed, we update
986 * both the local and the external one, and set it to
987 * the maximum of both */
995 f
->header
->tail_entry_seqnum
= htole64(r
);
997 if (f
->header
->head_entry_seqnum
== 0)
998 f
->header
->head_entry_seqnum
= htole64(r
);
1003 int journal_file_append_object(
1008 uint64_t *ret_offset
) {
1017 assert(type
> OBJECT_UNUSED
&& type
< _OBJECT_TYPE_MAX
);
1018 assert(size
>= sizeof(ObjectHeader
));
1020 r
= journal_file_set_online(f
);
1024 p
= le64toh(f
->header
->tail_object_offset
);
1026 p
= le64toh(f
->header
->header_size
);
1030 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &tail
);
1034 sz
= le64toh(READ_NOW(tail
->object
.size
));
1035 if (sz
> UINT64_MAX
- sizeof(uint64_t) + 1)
1039 if (p
> UINT64_MAX
- sz
)
1045 r
= journal_file_allocate(f
, p
, size
);
1049 r
= journal_file_move_to(f
, type
, false, p
, size
, &t
, NULL
);
1054 o
->object
= (ObjectHeader
) {
1056 .size
= htole64(size
),
1059 f
->header
->tail_object_offset
= htole64(p
);
1060 f
->header
->n_objects
= htole64(le64toh(f
->header
->n_objects
) + 1);
1071 static int journal_file_setup_data_hash_table(JournalFile
*f
) {
1079 /* We estimate that we need 1 hash table entry per 768 bytes
1080 of journal file and we want to make sure we never get
1081 beyond 75% fill level. Calculate the hash table size for
1082 the maximum file size based on these metrics. */
1084 s
= (f
->metrics
.max_size
* 4 / 768 / 3) * sizeof(HashItem
);
1085 if (s
< DEFAULT_DATA_HASH_TABLE_SIZE
)
1086 s
= DEFAULT_DATA_HASH_TABLE_SIZE
;
1088 log_debug("Reserving %"PRIu64
" entries in data hash table.", s
/ sizeof(HashItem
));
1090 r
= journal_file_append_object(f
,
1091 OBJECT_DATA_HASH_TABLE
,
1092 offsetof(Object
, hash_table
.items
) + s
,
1097 memzero(o
->hash_table
.items
, s
);
1099 f
->header
->data_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
1100 f
->header
->data_hash_table_size
= htole64(s
);
1105 static int journal_file_setup_field_hash_table(JournalFile
*f
) {
1113 /* We use a fixed size hash table for the fields as this
1114 * number should grow very slowly only */
1116 s
= DEFAULT_FIELD_HASH_TABLE_SIZE
;
1117 log_debug("Reserving %"PRIu64
" entries in field hash table.", s
/ sizeof(HashItem
));
1119 r
= journal_file_append_object(f
,
1120 OBJECT_FIELD_HASH_TABLE
,
1121 offsetof(Object
, hash_table
.items
) + s
,
1126 memzero(o
->hash_table
.items
, s
);
1128 f
->header
->field_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
1129 f
->header
->field_hash_table_size
= htole64(s
);
1134 int journal_file_map_data_hash_table(JournalFile
*f
) {
1142 if (f
->data_hash_table
)
1145 p
= le64toh(f
->header
->data_hash_table_offset
);
1146 s
= le64toh(f
->header
->data_hash_table_size
);
1148 r
= journal_file_move_to(f
,
1149 OBJECT_DATA_HASH_TABLE
,
1156 f
->data_hash_table
= t
;
1160 int journal_file_map_field_hash_table(JournalFile
*f
) {
1168 if (f
->field_hash_table
)
1171 p
= le64toh(f
->header
->field_hash_table_offset
);
1172 s
= le64toh(f
->header
->field_hash_table_size
);
1174 r
= journal_file_move_to(f
,
1175 OBJECT_FIELD_HASH_TABLE
,
1182 f
->field_hash_table
= t
;
1186 static int journal_file_link_field(
1197 assert(f
->field_hash_table
);
1201 if (o
->object
.type
!= OBJECT_FIELD
)
1204 m
= le64toh(READ_NOW(f
->header
->field_hash_table_size
)) / sizeof(HashItem
);
1208 /* This might alter the window we are looking at */
1209 o
->field
.next_hash_offset
= o
->field
.head_data_offset
= 0;
1212 p
= le64toh(f
->field_hash_table
[h
].tail_hash_offset
);
1214 f
->field_hash_table
[h
].head_hash_offset
= htole64(offset
);
1216 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1220 o
->field
.next_hash_offset
= htole64(offset
);
1223 f
->field_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1225 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
1226 f
->header
->n_fields
= htole64(le64toh(f
->header
->n_fields
) + 1);
1231 static int journal_file_link_data(
1242 assert(f
->data_hash_table
);
1246 if (o
->object
.type
!= OBJECT_DATA
)
1249 m
= le64toh(READ_NOW(f
->header
->data_hash_table_size
)) / sizeof(HashItem
);
1253 /* This might alter the window we are looking at */
1254 o
->data
.next_hash_offset
= o
->data
.next_field_offset
= 0;
1255 o
->data
.entry_offset
= o
->data
.entry_array_offset
= 0;
1256 o
->data
.n_entries
= 0;
1259 p
= le64toh(f
->data_hash_table
[h
].tail_hash_offset
);
1261 /* Only entry in the hash table is easy */
1262 f
->data_hash_table
[h
].head_hash_offset
= htole64(offset
);
1264 /* Move back to the previous data object, to patch in
1267 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1271 o
->data
.next_hash_offset
= htole64(offset
);
1274 f
->data_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1276 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
1277 f
->header
->n_data
= htole64(le64toh(f
->header
->n_data
) + 1);
1282 int journal_file_find_field_object_with_hash(
1284 const void *field
, uint64_t size
, uint64_t hash
,
1285 Object
**ret
, uint64_t *ret_offset
) {
1287 uint64_t p
, osize
, h
, m
;
1292 assert(field
&& size
> 0);
1294 /* If the field hash table is empty, we can't find anything */
1295 if (le64toh(f
->header
->field_hash_table_size
) <= 0)
1298 /* Map the field hash table, if it isn't mapped yet. */
1299 r
= journal_file_map_field_hash_table(f
);
1303 osize
= offsetof(Object
, field
.payload
) + size
;
1305 m
= le64toh(READ_NOW(f
->header
->field_hash_table_size
)) / sizeof(HashItem
);
1310 p
= le64toh(f
->field_hash_table
[h
].head_hash_offset
);
1315 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1319 if (le64toh(o
->field
.hash
) == hash
&&
1320 le64toh(o
->object
.size
) == osize
&&
1321 memcmp(o
->field
.payload
, field
, size
) == 0) {
1331 p
= le64toh(o
->field
.next_hash_offset
);
1337 int journal_file_find_field_object(
1339 const void *field
, uint64_t size
,
1340 Object
**ret
, uint64_t *ret_offset
) {
1345 assert(field
&& size
> 0);
1347 hash
= jenkins_hash64(field
, size
);
1349 return journal_file_find_field_object_with_hash(
1355 int journal_file_find_data_object_with_hash(
1357 const void *data
, uint64_t size
, uint64_t hash
,
1358 Object
**ret
, uint64_t *ret_offset
) {
1360 uint64_t p
, osize
, h
, m
;
1365 assert(data
|| size
== 0);
1367 /* If there's no data hash table, then there's no entry. */
1368 if (le64toh(f
->header
->data_hash_table_size
) <= 0)
1371 /* Map the data hash table, if it isn't mapped yet. */
1372 r
= journal_file_map_data_hash_table(f
);
1376 osize
= offsetof(Object
, data
.payload
) + size
;
1378 m
= le64toh(READ_NOW(f
->header
->data_hash_table_size
)) / sizeof(HashItem
);
1383 p
= le64toh(f
->data_hash_table
[h
].head_hash_offset
);
1388 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1392 if (le64toh(o
->data
.hash
) != hash
)
1395 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
) {
1396 #if HAVE_XZ || HAVE_LZ4
1400 l
= le64toh(READ_NOW(o
->object
.size
));
1401 if (l
<= offsetof(Object
, data
.payload
))
1404 l
-= offsetof(Object
, data
.payload
);
1406 r
= decompress_blob(o
->object
.flags
& OBJECT_COMPRESSION_MASK
,
1407 o
->data
.payload
, l
, &f
->compress_buffer
, &f
->compress_buffer_size
, &rsize
, 0);
1411 if (rsize
== size
&&
1412 memcmp(f
->compress_buffer
, data
, size
) == 0) {
1423 return -EPROTONOSUPPORT
;
1425 } else if (le64toh(o
->object
.size
) == osize
&&
1426 memcmp(o
->data
.payload
, data
, size
) == 0) {
1438 p
= le64toh(o
->data
.next_hash_offset
);
1444 int journal_file_find_data_object(
1446 const void *data
, uint64_t size
,
1447 Object
**ret
, uint64_t *ret_offset
) {
1452 assert(data
|| size
== 0);
1454 hash
= jenkins_hash64(data
, size
);
1456 return journal_file_find_data_object_with_hash(
1462 static int journal_file_append_field(
1464 const void *field
, uint64_t size
,
1465 Object
**ret
, uint64_t *ret_offset
) {
1473 assert(field
&& size
> 0);
1475 hash
= jenkins_hash64(field
, size
);
1477 r
= journal_file_find_field_object_with_hash(f
, field
, size
, hash
, &o
, &p
);
1491 osize
= offsetof(Object
, field
.payload
) + size
;
1492 r
= journal_file_append_object(f
, OBJECT_FIELD
, osize
, &o
, &p
);
1496 o
->field
.hash
= htole64(hash
);
1497 memcpy(o
->field
.payload
, field
, size
);
1499 r
= journal_file_link_field(f
, o
, p
, hash
);
1503 /* The linking might have altered the window, so let's
1504 * refresh our pointer */
1505 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1510 r
= journal_file_hmac_put_object(f
, OBJECT_FIELD
, o
, p
);
1524 static int journal_file_append_data(
1526 const void *data
, uint64_t size
,
1527 Object
**ret
, uint64_t *ret_offset
) {
1532 int r
, compression
= 0;
1536 assert(data
|| size
== 0);
1538 hash
= jenkins_hash64(data
, size
);
1540 r
= journal_file_find_data_object_with_hash(f
, data
, size
, hash
, &o
, &p
);
1554 osize
= offsetof(Object
, data
.payload
) + size
;
1555 r
= journal_file_append_object(f
, OBJECT_DATA
, osize
, &o
, &p
);
1559 o
->data
.hash
= htole64(hash
);
1561 #if HAVE_XZ || HAVE_LZ4
1562 if (JOURNAL_FILE_COMPRESS(f
) && size
>= f
->compress_threshold_bytes
) {
1565 compression
= compress_blob(data
, size
, o
->data
.payload
, size
- 1, &rsize
);
1567 if (compression
>= 0) {
1568 o
->object
.size
= htole64(offsetof(Object
, data
.payload
) + rsize
);
1569 o
->object
.flags
|= compression
;
1571 log_debug("Compressed data object %"PRIu64
" -> %zu using %s",
1572 size
, rsize
, object_compressed_to_string(compression
));
1574 /* Compression didn't work, we don't really care why, let's continue without compression */
1579 if (compression
== 0)
1580 memcpy_safe(o
->data
.payload
, data
, size
);
1582 r
= journal_file_link_data(f
, o
, p
, hash
);
1587 r
= journal_file_hmac_put_object(f
, OBJECT_DATA
, o
, p
);
1592 /* The linking might have altered the window, so let's
1593 * refresh our pointer */
1594 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1601 eq
= memchr(data
, '=', size
);
1602 if (eq
&& eq
> data
) {
1606 /* Create field object ... */
1607 r
= journal_file_append_field(f
, data
, (uint8_t*) eq
- (uint8_t*) data
, &fo
, &fp
);
1611 /* ... and link it in. */
1612 o
->data
.next_field_offset
= fo
->field
.head_data_offset
;
1613 fo
->field
.head_data_offset
= le64toh(p
);
1625 uint64_t journal_file_entry_n_items(Object
*o
) {
1629 if (o
->object
.type
!= OBJECT_ENTRY
)
1632 sz
= le64toh(READ_NOW(o
->object
.size
));
1633 if (sz
< offsetof(Object
, entry
.items
))
1636 return (sz
- offsetof(Object
, entry
.items
)) / sizeof(EntryItem
);
1639 uint64_t journal_file_entry_array_n_items(Object
*o
) {
1644 if (o
->object
.type
!= OBJECT_ENTRY_ARRAY
)
1647 sz
= le64toh(READ_NOW(o
->object
.size
));
1648 if (sz
< offsetof(Object
, entry_array
.items
))
1651 return (sz
- offsetof(Object
, entry_array
.items
)) / sizeof(uint64_t);
1654 uint64_t journal_file_hash_table_n_items(Object
*o
) {
1659 if (!IN_SET(o
->object
.type
, OBJECT_DATA_HASH_TABLE
, OBJECT_FIELD_HASH_TABLE
))
1662 sz
= le64toh(READ_NOW(o
->object
.size
));
1663 if (sz
< offsetof(Object
, hash_table
.items
))
1666 return (sz
- offsetof(Object
, hash_table
.items
)) / sizeof(HashItem
);
1669 static int link_entry_into_array(JournalFile
*f
,
1674 uint64_t n
= 0, ap
= 0, q
, i
, a
, hidx
;
1683 a
= le64toh(*first
);
1684 i
= hidx
= le64toh(READ_NOW(*idx
));
1687 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
1691 n
= journal_file_entry_array_n_items(o
);
1693 o
->entry_array
.items
[i
] = htole64(p
);
1694 *idx
= htole64(hidx
+ 1);
1700 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
1711 r
= journal_file_append_object(f
, OBJECT_ENTRY_ARRAY
,
1712 offsetof(Object
, entry_array
.items
) + n
* sizeof(uint64_t),
1718 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY_ARRAY
, o
, q
);
1723 o
->entry_array
.items
[i
] = htole64(p
);
1726 *first
= htole64(q
);
1728 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, ap
, &o
);
1732 o
->entry_array
.next_entry_array_offset
= htole64(q
);
1735 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
1736 f
->header
->n_entry_arrays
= htole64(le64toh(f
->header
->n_entry_arrays
) + 1);
1738 *idx
= htole64(hidx
+ 1);
1743 static int link_entry_into_array_plus_one(JournalFile
*f
,
1758 hidx
= le64toh(READ_NOW(*idx
));
1759 if (hidx
== UINT64_MAX
)
1762 *extra
= htole64(p
);
1766 i
= htole64(hidx
- 1);
1767 r
= link_entry_into_array(f
, first
, &i
, p
);
1772 *idx
= htole64(hidx
+ 1);
1776 static int journal_file_link_entry_item(JournalFile
*f
, Object
*o
, uint64_t offset
, uint64_t i
) {
1784 p
= le64toh(o
->entry
.items
[i
].object_offset
);
1785 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1789 return link_entry_into_array_plus_one(f
,
1790 &o
->data
.entry_offset
,
1791 &o
->data
.entry_array_offset
,
1796 static int journal_file_link_entry(JournalFile
*f
, Object
*o
, uint64_t offset
) {
1805 if (o
->object
.type
!= OBJECT_ENTRY
)
1808 __sync_synchronize();
1810 /* Link up the entry itself */
1811 r
= link_entry_into_array(f
,
1812 &f
->header
->entry_array_offset
,
1813 &f
->header
->n_entries
,
1818 /* log_debug("=> %s seqnr=%"PRIu64" n_entries=%"PRIu64, f->path, o->entry.seqnum, f->header->n_entries); */
1820 if (f
->header
->head_entry_realtime
== 0)
1821 f
->header
->head_entry_realtime
= o
->entry
.realtime
;
1823 f
->header
->tail_entry_realtime
= o
->entry
.realtime
;
1824 f
->header
->tail_entry_monotonic
= o
->entry
.monotonic
;
1826 /* Link up the items */
1827 n
= journal_file_entry_n_items(o
);
1828 for (i
= 0; i
< n
; i
++) {
1829 r
= journal_file_link_entry_item(f
, o
, offset
, i
);
1837 static int journal_file_append_entry_internal(
1839 const dual_timestamp
*ts
,
1840 const sd_id128_t
*boot_id
,
1842 const EntryItem items
[], unsigned n_items
,
1844 Object
**ret
, uint64_t *ret_offset
) {
1852 assert(items
|| n_items
== 0);
1855 osize
= offsetof(Object
, entry
.items
) + (n_items
* sizeof(EntryItem
));
1857 r
= journal_file_append_object(f
, OBJECT_ENTRY
, osize
, &o
, &np
);
1861 o
->entry
.seqnum
= htole64(journal_file_entry_seqnum(f
, seqnum
));
1862 memcpy_safe(o
->entry
.items
, items
, n_items
* sizeof(EntryItem
));
1863 o
->entry
.realtime
= htole64(ts
->realtime
);
1864 o
->entry
.monotonic
= htole64(ts
->monotonic
);
1865 o
->entry
.xor_hash
= htole64(xor_hash
);
1867 f
->header
->boot_id
= *boot_id
;
1868 o
->entry
.boot_id
= f
->header
->boot_id
;
1871 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY
, o
, np
);
1876 r
= journal_file_link_entry(f
, o
, np
);
1889 void journal_file_post_change(JournalFile
*f
) {
1895 /* inotify() does not receive IN_MODIFY events from file
1896 * accesses done via mmap(). After each access we hence
1897 * trigger IN_MODIFY by truncating the journal file to its
1898 * current size which triggers IN_MODIFY. */
1900 __sync_synchronize();
1902 if (ftruncate(f
->fd
, f
->last_stat
.st_size
) < 0)
1903 log_debug_errno(errno
, "Failed to truncate file to its own size: %m");
1906 static int post_change_thunk(sd_event_source
*timer
, uint64_t usec
, void *userdata
) {
1909 journal_file_post_change(userdata
);
1914 static void schedule_post_change(JournalFile
*f
) {
1919 assert(f
->post_change_timer
);
1921 r
= sd_event_source_get_enabled(f
->post_change_timer
, NULL
);
1923 log_debug_errno(r
, "Failed to get ftruncate timer state: %m");
1929 r
= sd_event_now(sd_event_source_get_event(f
->post_change_timer
), CLOCK_MONOTONIC
, &now
);
1931 log_debug_errno(r
, "Failed to get clock's now for scheduling ftruncate: %m");
1935 r
= sd_event_source_set_time(f
->post_change_timer
, now
+ f
->post_change_timer_period
);
1937 log_debug_errno(r
, "Failed to set time for scheduling ftruncate: %m");
1941 r
= sd_event_source_set_enabled(f
->post_change_timer
, SD_EVENT_ONESHOT
);
1943 log_debug_errno(r
, "Failed to enable scheduled ftruncate: %m");
1950 /* On failure, let's simply post the change immediately. */
1951 journal_file_post_change(f
);
1954 /* Enable coalesced change posting in a timer on the provided sd_event instance */
1955 int journal_file_enable_post_change_timer(JournalFile
*f
, sd_event
*e
, usec_t t
) {
1956 _cleanup_(sd_event_source_unrefp
) sd_event_source
*timer
= NULL
;
1960 assert_return(!f
->post_change_timer
, -EINVAL
);
1964 r
= sd_event_add_time(e
, &timer
, CLOCK_MONOTONIC
, 0, 0, post_change_thunk
, f
);
1968 r
= sd_event_source_set_enabled(timer
, SD_EVENT_OFF
);
1972 f
->post_change_timer
= TAKE_PTR(timer
);
1973 f
->post_change_timer_period
= t
;
1978 static int entry_item_cmp(const EntryItem
*a
, const EntryItem
*b
) {
1979 return CMP(le64toh(a
->object_offset
), le64toh(b
->object_offset
));
1982 int journal_file_append_entry(
1984 const dual_timestamp
*ts
,
1985 const sd_id128_t
*boot_id
,
1986 const struct iovec iovec
[], unsigned n_iovec
,
1988 Object
**ret
, uint64_t *ret_offset
) {
1993 uint64_t xor_hash
= 0;
1994 struct dual_timestamp _ts
;
1998 assert(iovec
|| n_iovec
== 0);
2001 if (!VALID_REALTIME(ts
->realtime
))
2002 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
2003 "Invalid realtime timestamp %" PRIu64
", refusing entry.",
2005 if (!VALID_MONOTONIC(ts
->monotonic
))
2006 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
2007 "Invalid monotomic timestamp %" PRIu64
", refusing entry.",
2010 dual_timestamp_get(&_ts
);
2015 r
= journal_file_maybe_append_tag(f
, ts
->realtime
);
2020 /* alloca() can't take 0, hence let's allocate at least one */
2021 items
= newa(EntryItem
, MAX(1u, n_iovec
));
2023 for (i
= 0; i
< n_iovec
; i
++) {
2027 r
= journal_file_append_data(f
, iovec
[i
].iov_base
, iovec
[i
].iov_len
, &o
, &p
);
2031 xor_hash
^= le64toh(o
->data
.hash
);
2032 items
[i
].object_offset
= htole64(p
);
2033 items
[i
].hash
= o
->data
.hash
;
2036 /* Order by the position on disk, in order to improve seek
2037 * times for rotating media. */
2038 typesafe_qsort(items
, n_iovec
, entry_item_cmp
);
2040 r
= journal_file_append_entry_internal(f
, ts
, boot_id
, xor_hash
, items
, n_iovec
, seqnum
, ret
, ret_offset
);
2042 /* If the memory mapping triggered a SIGBUS then we return an
2043 * IO error and ignore the error code passed down to us, since
2044 * it is very likely just an effect of a nullified replacement
2047 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
2050 if (f
->post_change_timer
)
2051 schedule_post_change(f
);
2053 journal_file_post_change(f
);
2058 typedef struct ChainCacheItem
{
2059 uint64_t first
; /* the array at the beginning of the chain */
2060 uint64_t array
; /* the cached array */
2061 uint64_t begin
; /* the first item in the cached array */
2062 uint64_t total
; /* the total number of items in all arrays before this one in the chain */
2063 uint64_t last_index
; /* the last index we looked at, to optimize locality when bisecting */
2066 static void chain_cache_put(
2073 uint64_t last_index
) {
2076 /* If the chain item to cache for this chain is the
2077 * first one it's not worth caching anything */
2081 if (ordered_hashmap_size(h
) >= CHAIN_CACHE_MAX
) {
2082 ci
= ordered_hashmap_steal_first(h
);
2085 ci
= new(ChainCacheItem
, 1);
2092 if (ordered_hashmap_put(h
, &ci
->first
, ci
) < 0) {
2097 assert(ci
->first
== first
);
2102 ci
->last_index
= last_index
;
2105 static int generic_array_get(
2109 Object
**ret
, uint64_t *ret_offset
) {
2112 uint64_t p
= 0, a
, t
= 0;
2120 /* Try the chain cache first */
2121 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
2122 if (ci
&& i
> ci
->total
) {
2131 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
2135 k
= journal_file_entry_array_n_items(o
);
2137 p
= le64toh(o
->entry_array
.items
[i
]);
2143 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
2149 /* Let's cache this item for the next invocation */
2150 chain_cache_put(f
->chain_cache
, ci
, first
, a
, le64toh(o
->entry_array
.items
[0]), t
, i
);
2152 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2165 static int generic_array_get_plus_one(
2170 Object
**ret
, uint64_t *ret_offset
) {
2179 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, &o
);
2187 *ret_offset
= extra
;
2192 return generic_array_get(f
, first
, i
-1, ret
, ret_offset
);
2201 static int generic_array_bisect(
2206 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
2207 direction_t direction
,
2209 uint64_t *ret_offset
,
2210 uint64_t *ret_idx
) {
2212 uint64_t a
, p
, t
= 0, i
= 0, last_p
= 0, last_index
= (uint64_t) -1;
2213 bool subtract_one
= false;
2214 Object
*o
, *array
= NULL
;
2219 assert(test_object
);
2221 /* Start with the first array in the chain */
2224 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
2225 if (ci
&& n
> ci
->total
&& ci
->begin
!= 0) {
2226 /* Ah, we have iterated this bisection array chain
2227 * previously! Let's see if we can skip ahead in the
2228 * chain, as far as the last time. But we can't jump
2229 * backwards in the chain, so let's check that
2232 r
= test_object(f
, ci
->begin
, needle
);
2236 if (r
== TEST_LEFT
) {
2237 /* OK, what we are looking for is right of the
2238 * begin of this EntryArray, so let's jump
2239 * straight to previously cached array in the
2245 last_index
= ci
->last_index
;
2250 uint64_t left
, right
, k
, lp
;
2252 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &array
);
2256 k
= journal_file_entry_array_n_items(array
);
2262 lp
= p
= le64toh(array
->entry_array
.items
[i
]);
2266 r
= test_object(f
, p
, needle
);
2267 if (r
== -EBADMSG
) {
2268 log_debug_errno(r
, "Encountered invalid entry while bisecting, cutting algorithm short. (1)");
2275 if (r
== TEST_FOUND
)
2276 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2278 if (r
== TEST_RIGHT
) {
2282 if (last_index
!= (uint64_t) -1) {
2283 assert(last_index
<= right
);
2285 /* If we cached the last index we
2286 * looked at, let's try to not to jump
2287 * too wildly around and see if we can
2288 * limit the range to look at early to
2289 * the immediate neighbors of the last
2290 * index we looked at. */
2292 if (last_index
> 0) {
2293 uint64_t x
= last_index
- 1;
2295 p
= le64toh(array
->entry_array
.items
[x
]);
2299 r
= test_object(f
, p
, needle
);
2303 if (r
== TEST_FOUND
)
2304 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2306 if (r
== TEST_RIGHT
)
2312 if (last_index
< right
) {
2313 uint64_t y
= last_index
+ 1;
2315 p
= le64toh(array
->entry_array
.items
[y
]);
2319 r
= test_object(f
, p
, needle
);
2323 if (r
== TEST_FOUND
)
2324 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2326 if (r
== TEST_RIGHT
)
2334 if (left
== right
) {
2335 if (direction
== DIRECTION_UP
)
2336 subtract_one
= true;
2342 assert(left
< right
);
2343 i
= (left
+ right
) / 2;
2345 p
= le64toh(array
->entry_array
.items
[i
]);
2349 r
= test_object(f
, p
, needle
);
2350 if (r
== -EBADMSG
) {
2351 log_debug_errno(r
, "Encountered invalid entry while bisecting, cutting algorithm short. (2)");
2358 if (r
== TEST_FOUND
)
2359 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2361 if (r
== TEST_RIGHT
)
2369 if (direction
== DIRECTION_UP
) {
2371 subtract_one
= true;
2382 last_index
= (uint64_t) -1;
2383 a
= le64toh(array
->entry_array
.next_entry_array_offset
);
2389 if (subtract_one
&& t
== 0 && i
== 0)
2392 /* Let's cache this item for the next invocation */
2393 chain_cache_put(f
->chain_cache
, ci
, first
, a
, le64toh(array
->entry_array
.items
[0]), t
, subtract_one
? (i
> 0 ? i
-1 : (uint64_t) -1) : i
);
2395 if (subtract_one
&& i
== 0)
2397 else if (subtract_one
)
2398 p
= le64toh(array
->entry_array
.items
[i
-1]);
2400 p
= le64toh(array
->entry_array
.items
[i
]);
2402 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2413 *ret_idx
= t
+ i
+ (subtract_one
? -1 : 0);
2418 static int generic_array_bisect_plus_one(
2424 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
2425 direction_t direction
,
2427 uint64_t *ret_offset
,
2428 uint64_t *ret_idx
) {
2431 bool step_back
= false;
2435 assert(test_object
);
2440 /* This bisects the array in object 'first', but first checks
2442 r
= test_object(f
, extra
, needle
);
2446 if (r
== TEST_FOUND
)
2447 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2449 /* if we are looking with DIRECTION_UP then we need to first
2450 see if in the actual array there is a matching entry, and
2451 return the last one of that. But if there isn't any we need
2452 to return this one. Hence remember this, and return it
2455 step_back
= direction
== DIRECTION_UP
;
2457 if (r
== TEST_RIGHT
) {
2458 if (direction
== DIRECTION_DOWN
)
2464 r
= generic_array_bisect(f
, first
, n
-1, needle
, test_object
, direction
, ret
, ret_offset
, ret_idx
);
2466 if (r
== 0 && step_back
)
2469 if (r
> 0 && ret_idx
)
2475 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, &o
);
2483 *ret_offset
= extra
;
2491 _pure_
static int test_object_offset(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2497 else if (p
< needle
)
2503 static int test_object_seqnum(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2511 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2515 sq
= le64toh(READ_NOW(o
->entry
.seqnum
));
2518 else if (sq
< needle
)
2524 int journal_file_move_to_entry_by_seqnum(
2527 direction_t direction
,
2529 uint64_t *ret_offset
) {
2533 return generic_array_bisect(
2535 le64toh(f
->header
->entry_array_offset
),
2536 le64toh(f
->header
->n_entries
),
2540 ret
, ret_offset
, NULL
);
2543 static int test_object_realtime(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2551 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2555 rt
= le64toh(READ_NOW(o
->entry
.realtime
));
2558 else if (rt
< needle
)
2564 int journal_file_move_to_entry_by_realtime(
2567 direction_t direction
,
2569 uint64_t *ret_offset
) {
2573 return generic_array_bisect(
2575 le64toh(f
->header
->entry_array_offset
),
2576 le64toh(f
->header
->n_entries
),
2578 test_object_realtime
,
2580 ret
, ret_offset
, NULL
);
2583 static int test_object_monotonic(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2591 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2595 m
= le64toh(READ_NOW(o
->entry
.monotonic
));
2598 else if (m
< needle
)
2604 static int find_data_object_by_boot_id(
2610 char t
[STRLEN("_BOOT_ID=") + 32 + 1] = "_BOOT_ID=";
2612 sd_id128_to_string(boot_id
, t
+ 9);
2613 return journal_file_find_data_object(f
, t
, sizeof(t
) - 1, o
, b
);
2616 int journal_file_move_to_entry_by_monotonic(
2620 direction_t direction
,
2622 uint64_t *ret_offset
) {
2629 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, NULL
);
2635 return generic_array_bisect_plus_one(
2637 le64toh(o
->data
.entry_offset
),
2638 le64toh(o
->data
.entry_array_offset
),
2639 le64toh(o
->data
.n_entries
),
2641 test_object_monotonic
,
2643 ret
, ret_offset
, NULL
);
2646 void journal_file_reset_location(JournalFile
*f
) {
2647 f
->location_type
= LOCATION_HEAD
;
2648 f
->current_offset
= 0;
2649 f
->current_seqnum
= 0;
2650 f
->current_realtime
= 0;
2651 f
->current_monotonic
= 0;
2652 zero(f
->current_boot_id
);
2653 f
->current_xor_hash
= 0;
2656 void journal_file_save_location(JournalFile
*f
, Object
*o
, uint64_t offset
) {
2657 f
->location_type
= LOCATION_SEEK
;
2658 f
->current_offset
= offset
;
2659 f
->current_seqnum
= le64toh(o
->entry
.seqnum
);
2660 f
->current_realtime
= le64toh(o
->entry
.realtime
);
2661 f
->current_monotonic
= le64toh(o
->entry
.monotonic
);
2662 f
->current_boot_id
= o
->entry
.boot_id
;
2663 f
->current_xor_hash
= le64toh(o
->entry
.xor_hash
);
2666 int journal_file_compare_locations(JournalFile
*af
, JournalFile
*bf
) {
2673 assert(af
->location_type
== LOCATION_SEEK
);
2674 assert(bf
->location_type
== LOCATION_SEEK
);
2676 /* If contents and timestamps match, these entries are
2677 * identical, even if the seqnum does not match */
2678 if (sd_id128_equal(af
->current_boot_id
, bf
->current_boot_id
) &&
2679 af
->current_monotonic
== bf
->current_monotonic
&&
2680 af
->current_realtime
== bf
->current_realtime
&&
2681 af
->current_xor_hash
== bf
->current_xor_hash
)
2684 if (sd_id128_equal(af
->header
->seqnum_id
, bf
->header
->seqnum_id
)) {
2686 /* If this is from the same seqnum source, compare
2688 r
= CMP(af
->current_seqnum
, bf
->current_seqnum
);
2692 /* Wow! This is weird, different data but the same
2693 * seqnums? Something is borked, but let's make the
2694 * best of it and compare by time. */
2697 if (sd_id128_equal(af
->current_boot_id
, bf
->current_boot_id
)) {
2699 /* If the boot id matches, compare monotonic time */
2700 r
= CMP(af
->current_monotonic
, bf
->current_monotonic
);
2705 /* Otherwise, compare UTC time */
2706 r
= CMP(af
->current_realtime
, bf
->current_realtime
);
2710 /* Finally, compare by contents */
2711 return CMP(af
->current_xor_hash
, bf
->current_xor_hash
);
2714 static int bump_array_index(uint64_t *i
, direction_t direction
, uint64_t n
) {
2716 /* Increase or decrease the specified index, in the right direction. */
2718 if (direction
== DIRECTION_DOWN
) {
2733 static bool check_properly_ordered(uint64_t new_offset
, uint64_t old_offset
, direction_t direction
) {
2735 /* Consider it an error if any of the two offsets is uninitialized */
2736 if (old_offset
== 0 || new_offset
== 0)
2739 /* If we go down, the new offset must be larger than the old one. */
2740 return direction
== DIRECTION_DOWN
?
2741 new_offset
> old_offset
:
2742 new_offset
< old_offset
;
2745 int journal_file_next_entry(
2748 direction_t direction
,
2749 Object
**ret
, uint64_t *ret_offset
) {
2757 n
= le64toh(READ_NOW(f
->header
->n_entries
));
2762 i
= direction
== DIRECTION_DOWN
? 0 : n
- 1;
2764 r
= generic_array_bisect(f
,
2765 le64toh(f
->header
->entry_array_offset
),
2766 le64toh(f
->header
->n_entries
),
2775 r
= bump_array_index(&i
, direction
, n
);
2780 /* And jump to it */
2782 r
= generic_array_get(f
,
2783 le64toh(f
->header
->entry_array_offset
),
2791 /* OK, so this entry is borked. Most likely some entry didn't get synced to disk properly, let's see if
2792 * the next one might work for us instead. */
2793 log_debug_errno(r
, "Entry item %" PRIu64
" is bad, skipping over it.", i
);
2795 r
= bump_array_index(&i
, direction
, n
);
2800 /* Ensure our array is properly ordered. */
2801 if (p
> 0 && !check_properly_ordered(ofs
, p
, direction
))
2802 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
2803 "%s: entry array not properly ordered at entry %" PRIu64
,
2812 int journal_file_next_entry_for_data(
2814 Object
*o
, uint64_t p
,
2815 uint64_t data_offset
,
2816 direction_t direction
,
2817 Object
**ret
, uint64_t *ret_offset
) {
2824 assert(p
> 0 || !o
);
2826 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2830 n
= le64toh(READ_NOW(d
->data
.n_entries
));
2835 i
= direction
== DIRECTION_DOWN
? 0 : n
- 1;
2837 if (o
->object
.type
!= OBJECT_ENTRY
)
2840 r
= generic_array_bisect_plus_one(f
,
2841 le64toh(d
->data
.entry_offset
),
2842 le64toh(d
->data
.entry_array_offset
),
2843 le64toh(d
->data
.n_entries
),
2853 r
= bump_array_index(&i
, direction
, n
);
2859 r
= generic_array_get_plus_one(f
,
2860 le64toh(d
->data
.entry_offset
),
2861 le64toh(d
->data
.entry_array_offset
),
2869 log_debug_errno(r
, "Data entry item %" PRIu64
" is bad, skipping over it.", i
);
2871 r
= bump_array_index(&i
, direction
, n
);
2876 /* Ensure our array is properly ordered. */
2877 if (p
> 0 && check_properly_ordered(ofs
, p
, direction
))
2878 return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG
),
2879 "%s data entry array not properly ordered at entry %" PRIu64
,
2888 int journal_file_move_to_entry_by_offset_for_data(
2890 uint64_t data_offset
,
2892 direction_t direction
,
2893 Object
**ret
, uint64_t *ret_offset
) {
2900 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2904 return generic_array_bisect_plus_one(
2906 le64toh(d
->data
.entry_offset
),
2907 le64toh(d
->data
.entry_array_offset
),
2908 le64toh(d
->data
.n_entries
),
2912 ret
, ret_offset
, NULL
);
2915 int journal_file_move_to_entry_by_monotonic_for_data(
2917 uint64_t data_offset
,
2920 direction_t direction
,
2921 Object
**ret
, uint64_t *ret_offset
) {
2929 /* First, seek by time */
2930 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, &b
);
2936 r
= generic_array_bisect_plus_one(f
,
2937 le64toh(o
->data
.entry_offset
),
2938 le64toh(o
->data
.entry_array_offset
),
2939 le64toh(o
->data
.n_entries
),
2941 test_object_monotonic
,
2947 /* And now, continue seeking until we find an entry that
2948 * exists in both bisection arrays */
2954 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2958 r
= generic_array_bisect_plus_one(f
,
2959 le64toh(d
->data
.entry_offset
),
2960 le64toh(d
->data
.entry_array_offset
),
2961 le64toh(d
->data
.n_entries
),
2969 r
= journal_file_move_to_object(f
, OBJECT_DATA
, b
, &o
);
2973 r
= generic_array_bisect_plus_one(f
,
2974 le64toh(o
->data
.entry_offset
),
2975 le64toh(o
->data
.entry_array_offset
),
2976 le64toh(o
->data
.n_entries
),
2998 int journal_file_move_to_entry_by_seqnum_for_data(
3000 uint64_t data_offset
,
3002 direction_t direction
,
3003 Object
**ret
, uint64_t *ret_offset
) {
3010 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
3014 return generic_array_bisect_plus_one(
3016 le64toh(d
->data
.entry_offset
),
3017 le64toh(d
->data
.entry_array_offset
),
3018 le64toh(d
->data
.n_entries
),
3022 ret
, ret_offset
, NULL
);
3025 int journal_file_move_to_entry_by_realtime_for_data(
3027 uint64_t data_offset
,
3029 direction_t direction
,
3030 Object
**ret
, uint64_t *ret_offset
) {
3037 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
3041 return generic_array_bisect_plus_one(
3043 le64toh(d
->data
.entry_offset
),
3044 le64toh(d
->data
.entry_array_offset
),
3045 le64toh(d
->data
.n_entries
),
3047 test_object_realtime
,
3049 ret
, ret_offset
, NULL
);
3052 void journal_file_dump(JournalFile
*f
) {
3060 journal_file_print_header(f
);
3062 p
= le64toh(READ_NOW(f
->header
->header_size
));
3064 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &o
);
3068 switch (o
->object
.type
) {
3071 printf("Type: OBJECT_UNUSED\n");
3075 printf("Type: OBJECT_DATA\n");
3079 printf("Type: OBJECT_FIELD\n");
3083 printf("Type: OBJECT_ENTRY seqnum=%"PRIu64
" monotonic=%"PRIu64
" realtime=%"PRIu64
"\n",
3084 le64toh(o
->entry
.seqnum
),
3085 le64toh(o
->entry
.monotonic
),
3086 le64toh(o
->entry
.realtime
));
3089 case OBJECT_FIELD_HASH_TABLE
:
3090 printf("Type: OBJECT_FIELD_HASH_TABLE\n");
3093 case OBJECT_DATA_HASH_TABLE
:
3094 printf("Type: OBJECT_DATA_HASH_TABLE\n");
3097 case OBJECT_ENTRY_ARRAY
:
3098 printf("Type: OBJECT_ENTRY_ARRAY\n");
3102 printf("Type: OBJECT_TAG seqnum=%"PRIu64
" epoch=%"PRIu64
"\n",
3103 le64toh(o
->tag
.seqnum
),
3104 le64toh(o
->tag
.epoch
));
3108 printf("Type: unknown (%i)\n", o
->object
.type
);
3112 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
)
3113 printf("Flags: %s\n",
3114 object_compressed_to_string(o
->object
.flags
& OBJECT_COMPRESSION_MASK
));
3116 if (p
== le64toh(f
->header
->tail_object_offset
))
3119 p
+= ALIGN64(le64toh(o
->object
.size
));
3124 log_error("File corrupt");
3127 static const char* format_timestamp_safe(char *buf
, size_t l
, usec_t t
) {
3130 x
= format_timestamp(buf
, l
, t
);
3136 void journal_file_print_header(JournalFile
*f
) {
3137 char a
[SD_ID128_STRING_MAX
], b
[SD_ID128_STRING_MAX
], c
[SD_ID128_STRING_MAX
], d
[SD_ID128_STRING_MAX
];
3138 char x
[FORMAT_TIMESTAMP_MAX
], y
[FORMAT_TIMESTAMP_MAX
], z
[FORMAT_TIMESTAMP_MAX
];
3140 char bytes
[FORMAT_BYTES_MAX
];
3145 printf("File path: %s\n"
3149 "Sequential number ID: %s\n"
3151 "Compatible flags:%s%s\n"
3152 "Incompatible flags:%s%s%s\n"
3153 "Header size: %"PRIu64
"\n"
3154 "Arena size: %"PRIu64
"\n"
3155 "Data hash table size: %"PRIu64
"\n"
3156 "Field hash table size: %"PRIu64
"\n"
3157 "Rotate suggested: %s\n"
3158 "Head sequential number: %"PRIu64
" (%"PRIx64
")\n"
3159 "Tail sequential number: %"PRIu64
" (%"PRIx64
")\n"
3160 "Head realtime timestamp: %s (%"PRIx64
")\n"
3161 "Tail realtime timestamp: %s (%"PRIx64
")\n"
3162 "Tail monotonic timestamp: %s (%"PRIx64
")\n"
3163 "Objects: %"PRIu64
"\n"
3164 "Entry objects: %"PRIu64
"\n",
3166 sd_id128_to_string(f
->header
->file_id
, a
),
3167 sd_id128_to_string(f
->header
->machine_id
, b
),
3168 sd_id128_to_string(f
->header
->boot_id
, c
),
3169 sd_id128_to_string(f
->header
->seqnum_id
, d
),
3170 f
->header
->state
== STATE_OFFLINE
? "OFFLINE" :
3171 f
->header
->state
== STATE_ONLINE
? "ONLINE" :
3172 f
->header
->state
== STATE_ARCHIVED
? "ARCHIVED" : "UNKNOWN",
3173 JOURNAL_HEADER_SEALED(f
->header
) ? " SEALED" : "",
3174 (le32toh(f
->header
->compatible_flags
) & ~HEADER_COMPATIBLE_ANY
) ? " ???" : "",
3175 JOURNAL_HEADER_COMPRESSED_XZ(f
->header
) ? " COMPRESSED-XZ" : "",
3176 JOURNAL_HEADER_COMPRESSED_LZ4(f
->header
) ? " COMPRESSED-LZ4" : "",
3177 (le32toh(f
->header
->incompatible_flags
) & ~HEADER_INCOMPATIBLE_ANY
) ? " ???" : "",
3178 le64toh(f
->header
->header_size
),
3179 le64toh(f
->header
->arena_size
),
3180 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
3181 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
),
3182 yes_no(journal_file_rotate_suggested(f
, 0)),
3183 le64toh(f
->header
->head_entry_seqnum
), le64toh(f
->header
->head_entry_seqnum
),
3184 le64toh(f
->header
->tail_entry_seqnum
), le64toh(f
->header
->tail_entry_seqnum
),
3185 format_timestamp_safe(x
, sizeof(x
), le64toh(f
->header
->head_entry_realtime
)), le64toh(f
->header
->head_entry_realtime
),
3186 format_timestamp_safe(y
, sizeof(y
), le64toh(f
->header
->tail_entry_realtime
)), le64toh(f
->header
->tail_entry_realtime
),
3187 format_timespan(z
, sizeof(z
), le64toh(f
->header
->tail_entry_monotonic
), USEC_PER_MSEC
), le64toh(f
->header
->tail_entry_monotonic
),
3188 le64toh(f
->header
->n_objects
),
3189 le64toh(f
->header
->n_entries
));
3191 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
3192 printf("Data objects: %"PRIu64
"\n"
3193 "Data hash table fill: %.1f%%\n",
3194 le64toh(f
->header
->n_data
),
3195 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))));
3197 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
3198 printf("Field objects: %"PRIu64
"\n"
3199 "Field hash table fill: %.1f%%\n",
3200 le64toh(f
->header
->n_fields
),
3201 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))));
3203 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_tags
))
3204 printf("Tag objects: %"PRIu64
"\n",
3205 le64toh(f
->header
->n_tags
));
3206 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
3207 printf("Entry array objects: %"PRIu64
"\n",
3208 le64toh(f
->header
->n_entry_arrays
));
3210 if (fstat(f
->fd
, &st
) >= 0)
3211 printf("Disk usage: %s\n", format_bytes(bytes
, sizeof(bytes
), (uint64_t) st
.st_blocks
* 512ULL));
3214 static int journal_file_warn_btrfs(JournalFile
*f
) {
3220 /* Before we write anything, check if the COW logic is turned
3221 * off on btrfs. Given our write pattern that is quite
3222 * unfriendly to COW file systems this should greatly improve
3223 * performance on COW file systems, such as btrfs, at the
3224 * expense of data integrity features (which shouldn't be too
3225 * bad, given that we do our own checksumming). */
3227 r
= btrfs_is_filesystem(f
->fd
);
3229 return log_warning_errno(r
, "Failed to determine if journal is on btrfs: %m");
3233 r
= read_attr_fd(f
->fd
, &attrs
);
3235 return log_warning_errno(r
, "Failed to read file attributes: %m");
3237 if (attrs
& FS_NOCOW_FL
) {
3238 log_debug("Detected btrfs file system with copy-on-write disabled, all is good.");
3242 log_notice("Creating journal file %s on a btrfs file system, and copy-on-write is enabled. "
3243 "This is likely to slow down journal access substantially, please consider turning "
3244 "off the copy-on-write file attribute on the journal directory, using chattr +C.", f
->path
);
3249 int journal_file_open(
3255 uint64_t compress_threshold_bytes
,
3257 JournalMetrics
*metrics
,
3258 MMapCache
*mmap_cache
,
3259 Set
*deferred_closes
,
3260 JournalFile
*template,
3261 JournalFile
**ret
) {
3263 bool newly_created
= false;
3269 assert(fd
>= 0 || fname
);
3271 if (!IN_SET((flags
& O_ACCMODE
), O_RDONLY
, O_RDWR
))
3274 if (fname
&& (flags
& O_CREAT
) && !endswith(fname
, ".journal"))
3277 f
= new(JournalFile
, 1);
3281 *f
= (JournalFile
) {
3286 .prot
= prot_from_flags(flags
),
3287 .writable
= (flags
& O_ACCMODE
) != O_RDONLY
,
3290 .compress_lz4
= compress
,
3292 .compress_xz
= compress
,
3294 .compress_threshold_bytes
= compress_threshold_bytes
== (uint64_t) -1 ?
3295 DEFAULT_COMPRESS_THRESHOLD
:
3296 MAX(MIN_COMPRESS_THRESHOLD
, compress_threshold_bytes
),
3302 if (DEBUG_LOGGING
) {
3303 static int last_seal
= -1, last_compress
= -1;
3304 static uint64_t last_bytes
= UINT64_MAX
;
3305 char bytes
[FORMAT_BYTES_MAX
];
3307 if (last_seal
!= f
->seal
||
3308 last_compress
!= JOURNAL_FILE_COMPRESS(f
) ||
3309 last_bytes
!= f
->compress_threshold_bytes
) {
3311 log_debug("Journal effective settings seal=%s compress=%s compress_threshold_bytes=%s",
3312 yes_no(f
->seal
), yes_no(JOURNAL_FILE_COMPRESS(f
)),
3313 format_bytes(bytes
, sizeof bytes
, f
->compress_threshold_bytes
));
3314 last_seal
= f
->seal
;
3315 last_compress
= JOURNAL_FILE_COMPRESS(f
);
3316 last_bytes
= f
->compress_threshold_bytes
;
3321 f
->mmap
= mmap_cache_ref(mmap_cache
);
3323 f
->mmap
= mmap_cache_new();
3331 f
->path
= strdup(fname
);
3339 /* If we don't know the path, fill in something explanatory and vaguely useful */
3340 if (asprintf(&f
->path
, "/proc/self/%i", fd
) < 0) {
3346 f
->chain_cache
= ordered_hashmap_new(&uint64_hash_ops
);
3347 if (!f
->chain_cache
) {
3353 /* We pass O_NONBLOCK here, so that in case somebody pointed us to some character device node or FIFO
3354 * or so, we likely fail quickly than block for long. For regular files O_NONBLOCK has no effect, hence
3355 * it doesn't hurt in that case. */
3357 f
->fd
= open(f
->path
, f
->flags
|O_CLOEXEC
|O_NONBLOCK
, f
->mode
);
3363 /* fds we opened here by us should also be closed by us. */
3366 r
= fd_nonblock(f
->fd
, false);
3371 f
->cache_fd
= mmap_cache_add_fd(f
->mmap
, f
->fd
);
3377 r
= journal_file_fstat(f
);
3381 if (f
->last_stat
.st_size
== 0 && f
->writable
) {
3383 (void) journal_file_warn_btrfs(f
);
3385 /* Let's attach the creation time to the journal file, so that the vacuuming code knows the age of this
3386 * file even if the file might end up corrupted one day... Ideally we'd just use the creation time many
3387 * file systems maintain for each file, but the API to query this is very new, hence let's emulate this
3388 * via extended attributes. If extended attributes are not supported we'll just skip this, and rely
3389 * solely on mtime/atime/ctime of the file. */
3390 (void) fd_setcrtime(f
->fd
, 0);
3393 /* Try to load the FSPRG state, and if we can't, then
3394 * just don't do sealing */
3396 r
= journal_file_fss_load(f
);
3402 r
= journal_file_init_header(f
, template);
3406 r
= journal_file_fstat(f
);
3410 newly_created
= true;
3413 if (f
->last_stat
.st_size
< (off_t
) HEADER_SIZE_MIN
) {
3418 r
= mmap_cache_get(f
->mmap
, f
->cache_fd
, f
->prot
, CONTEXT_HEADER
, true, 0, PAGE_ALIGN(sizeof(Header
)), &f
->last_stat
, &h
, NULL
);
3420 /* Some file systems (jffs2 or p9fs) don't support mmap() properly (or only read-only
3421 * mmap()), and return EINVAL in that case. Let's propagate that as a more recognizable error
3431 if (!newly_created
) {
3432 set_clear_with_destructor(deferred_closes
, journal_file_close
);
3434 r
= journal_file_verify_header(f
);
3440 if (!newly_created
&& f
->writable
) {
3441 r
= journal_file_fss_load(f
);
3449 journal_default_metrics(metrics
, f
->fd
);
3450 f
->metrics
= *metrics
;
3451 } else if (template)
3452 f
->metrics
= template->metrics
;
3454 r
= journal_file_refresh_header(f
);
3460 r
= journal_file_hmac_setup(f
);
3465 if (newly_created
) {
3466 r
= journal_file_setup_field_hash_table(f
);
3470 r
= journal_file_setup_data_hash_table(f
);
3475 r
= journal_file_append_first_tag(f
);
3481 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
)) {
3486 if (template && template->post_change_timer
) {
3487 r
= journal_file_enable_post_change_timer(
3489 sd_event_source_get_event(template->post_change_timer
),
3490 template->post_change_timer_period
);
3496 /* The file is opened now successfully, thus we take possession of any passed in fd. */
3503 if (f
->cache_fd
&& mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
3506 (void) journal_file_close(f
);
3511 int journal_file_archive(JournalFile
*f
) {
3512 _cleanup_free_
char *p
= NULL
;
3519 /* Is this a journal file that was passed to us as fd? If so, we synthesized a path name for it, and we refuse
3520 * rotation, since we don't know the actual path, and couldn't rename the file hence. */
3521 if (path_startswith(f
->path
, "/proc/self/fd"))
3524 if (!endswith(f
->path
, ".journal"))
3527 if (asprintf(&p
, "%.*s@" SD_ID128_FORMAT_STR
"-%016"PRIx64
"-%016"PRIx64
".journal",
3528 (int) strlen(f
->path
) - 8, f
->path
,
3529 SD_ID128_FORMAT_VAL(f
->header
->seqnum_id
),
3530 le64toh(f
->header
->head_entry_seqnum
),
3531 le64toh(f
->header
->head_entry_realtime
)) < 0)
3534 /* Try to rename the file to the archived version. If the file already was deleted, we'll get ENOENT, let's
3535 * ignore that case. */
3536 if (rename(f
->path
, p
) < 0 && errno
!= ENOENT
)
3539 /* Sync the rename to disk */
3540 (void) fsync_directory_of_file(f
->fd
);
3542 /* Set as archive so offlining commits w/state=STATE_ARCHIVED. Previously we would set old_file->header->state
3543 * to STATE_ARCHIVED directly here, but journal_file_set_offline() short-circuits when state != STATE_ONLINE,
3544 * which would result in the rotated journal never getting fsync() called before closing. Now we simply queue
3545 * the archive state by setting an archive bit, leaving the state as STATE_ONLINE so proper offlining
3549 /* Currently, btrfs is not very good with out write patterns and fragments heavily. Let's defrag our journal
3550 * files when we archive them */
3551 f
->defrag_on_close
= true;
3556 JournalFile
* journal_initiate_close(
3558 Set
*deferred_closes
) {
3564 if (deferred_closes
) {
3566 r
= set_put(deferred_closes
, f
);
3568 log_debug_errno(r
, "Failed to add file to deferred close set, closing immediately.");
3570 (void) journal_file_set_offline(f
, false);
3575 return journal_file_close(f
);
3578 int journal_file_rotate(
3581 uint64_t compress_threshold_bytes
,
3583 Set
*deferred_closes
) {
3585 JournalFile
*new_file
= NULL
;
3591 r
= journal_file_archive(*f
);
3595 r
= journal_file_open(
3601 compress_threshold_bytes
,
3609 journal_initiate_close(*f
, deferred_closes
);
3615 int journal_file_dispose(int dir_fd
, const char *fname
) {
3616 _cleanup_free_
char *p
= NULL
;
3617 _cleanup_close_
int fd
= -1;
3621 /* Renames a journal file to *.journal~, i.e. to mark it as corruped or otherwise uncleanly shutdown. Note that
3622 * this is done without looking into the file or changing any of its contents. The idea is that this is called
3623 * whenever something is suspicious and we want to move the file away and make clear that it is not accessed
3624 * for writing anymore. */
3626 if (!endswith(fname
, ".journal"))
3629 if (asprintf(&p
, "%.*s@%016" PRIx64
"-%016" PRIx64
".journal~",
3630 (int) strlen(fname
) - 8, fname
,
3631 now(CLOCK_REALTIME
),
3635 if (renameat(dir_fd
, fname
, dir_fd
, p
) < 0)
3638 /* btrfs doesn't cope well with our write pattern and fragments heavily. Let's defrag all files we rotate */
3639 fd
= openat(dir_fd
, p
, O_RDONLY
|O_CLOEXEC
|O_NOCTTY
|O_NOFOLLOW
);
3641 log_debug_errno(errno
, "Failed to open file for defragmentation/FS_NOCOW_FL, ignoring: %m");
3643 (void) chattr_fd(fd
, 0, FS_NOCOW_FL
, NULL
);
3644 (void) btrfs_defrag_fd(fd
);
3650 int journal_file_open_reliably(
3655 uint64_t compress_threshold_bytes
,
3657 JournalMetrics
*metrics
,
3658 MMapCache
*mmap_cache
,
3659 Set
*deferred_closes
,
3660 JournalFile
*template,
3661 JournalFile
**ret
) {
3665 r
= journal_file_open(-1, fname
, flags
, mode
, compress
, compress_threshold_bytes
, seal
, metrics
, mmap_cache
,
3666 deferred_closes
, template, ret
);
3668 -EBADMSG
, /* Corrupted */
3669 -ENODATA
, /* Truncated */
3670 -EHOSTDOWN
, /* Other machine */
3671 -EPROTONOSUPPORT
, /* Incompatible feature */
3672 -EBUSY
, /* Unclean shutdown */
3673 -ESHUTDOWN
, /* Already archived */
3674 -EIO
, /* IO error, including SIGBUS on mmap */
3675 -EIDRM
, /* File has been deleted */
3676 -ETXTBSY
)) /* File is from the future */
3679 if ((flags
& O_ACCMODE
) == O_RDONLY
)
3682 if (!(flags
& O_CREAT
))
3685 if (!endswith(fname
, ".journal"))
3688 /* The file is corrupted. Rotate it away and try it again (but only once) */
3689 log_warning_errno(r
, "File %s corrupted or uncleanly shut down, renaming and replacing.", fname
);
3691 r
= journal_file_dispose(AT_FDCWD
, fname
);
3695 return journal_file_open(-1, fname
, flags
, mode
, compress
, compress_threshold_bytes
, seal
, metrics
, mmap_cache
,
3696 deferred_closes
, template, ret
);
3699 int journal_file_copy_entry(JournalFile
*from
, JournalFile
*to
, Object
*o
, uint64_t p
) {
3701 uint64_t q
, xor_hash
= 0;
3705 const sd_id128_t
*boot_id
;
3715 ts
.monotonic
= le64toh(o
->entry
.monotonic
);
3716 ts
.realtime
= le64toh(o
->entry
.realtime
);
3717 boot_id
= &o
->entry
.boot_id
;
3719 n
= journal_file_entry_n_items(o
);
3720 /* alloca() can't take 0, hence let's allocate at least one */
3721 items
= newa(EntryItem
, MAX(1u, n
));
3723 for (i
= 0; i
< n
; i
++) {
3730 q
= le64toh(o
->entry
.items
[i
].object_offset
);
3731 le_hash
= o
->entry
.items
[i
].hash
;
3733 r
= journal_file_move_to_object(from
, OBJECT_DATA
, q
, &o
);
3737 if (le_hash
!= o
->data
.hash
)
3740 l
= le64toh(READ_NOW(o
->object
.size
));
3741 if (l
< offsetof(Object
, data
.payload
))
3744 l
-= offsetof(Object
, data
.payload
);
3747 /* We hit the limit on 32bit machines */
3748 if ((uint64_t) t
!= l
)
3751 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
) {
3752 #if HAVE_XZ || HAVE_LZ4
3755 r
= decompress_blob(o
->object
.flags
& OBJECT_COMPRESSION_MASK
,
3756 o
->data
.payload
, l
, &from
->compress_buffer
, &from
->compress_buffer_size
, &rsize
, 0);
3760 data
= from
->compress_buffer
;
3763 return -EPROTONOSUPPORT
;
3766 data
= o
->data
.payload
;
3768 r
= journal_file_append_data(to
, data
, l
, &u
, &h
);
3772 xor_hash
^= le64toh(u
->data
.hash
);
3773 items
[i
].object_offset
= htole64(h
);
3774 items
[i
].hash
= u
->data
.hash
;
3776 r
= journal_file_move_to_object(from
, OBJECT_ENTRY
, p
, &o
);
3781 r
= journal_file_append_entry_internal(to
, &ts
, boot_id
, xor_hash
, items
, n
,
3784 if (mmap_cache_got_sigbus(to
->mmap
, to
->cache_fd
))
3790 void journal_reset_metrics(JournalMetrics
*m
) {
3793 /* Set everything to "pick automatic values". */
3795 *m
= (JournalMetrics
) {
3796 .min_use
= (uint64_t) -1,
3797 .max_use
= (uint64_t) -1,
3798 .min_size
= (uint64_t) -1,
3799 .max_size
= (uint64_t) -1,
3800 .keep_free
= (uint64_t) -1,
3801 .n_max_files
= (uint64_t) -1,
3805 void journal_default_metrics(JournalMetrics
*m
, int fd
) {
3806 char a
[FORMAT_BYTES_MAX
], b
[FORMAT_BYTES_MAX
], c
[FORMAT_BYTES_MAX
], d
[FORMAT_BYTES_MAX
], e
[FORMAT_BYTES_MAX
];
3808 uint64_t fs_size
= 0;
3813 if (fstatvfs(fd
, &ss
) >= 0)
3814 fs_size
= ss
.f_frsize
* ss
.f_blocks
;
3816 log_debug_errno(errno
, "Failed to determine disk size: %m");
3818 if (m
->max_use
== (uint64_t) -1) {
3821 m
->max_use
= CLAMP(PAGE_ALIGN(fs_size
/ 10), /* 10% of file system size */
3822 MAX_USE_LOWER
, MAX_USE_UPPER
);
3824 m
->max_use
= MAX_USE_LOWER
;
3826 m
->max_use
= PAGE_ALIGN(m
->max_use
);
3828 if (m
->max_use
!= 0 && m
->max_use
< JOURNAL_FILE_SIZE_MIN
*2)
3829 m
->max_use
= JOURNAL_FILE_SIZE_MIN
*2;
3832 if (m
->min_use
== (uint64_t) -1) {
3834 m
->min_use
= CLAMP(PAGE_ALIGN(fs_size
/ 50), /* 2% of file system size */
3835 MIN_USE_LOW
, MIN_USE_HIGH
);
3837 m
->min_use
= MIN_USE_LOW
;
3840 if (m
->min_use
> m
->max_use
)
3841 m
->min_use
= m
->max_use
;
3843 if (m
->max_size
== (uint64_t) -1)
3844 m
->max_size
= MIN(PAGE_ALIGN(m
->max_use
/ 8), /* 8 chunks */
3847 m
->max_size
= PAGE_ALIGN(m
->max_size
);
3849 if (m
->max_size
!= 0) {
3850 if (m
->max_size
< JOURNAL_FILE_SIZE_MIN
)
3851 m
->max_size
= JOURNAL_FILE_SIZE_MIN
;
3853 if (m
->max_use
!= 0 && m
->max_size
*2 > m
->max_use
)
3854 m
->max_use
= m
->max_size
*2;
3857 if (m
->min_size
== (uint64_t) -1)
3858 m
->min_size
= JOURNAL_FILE_SIZE_MIN
;
3860 m
->min_size
= CLAMP(PAGE_ALIGN(m
->min_size
),
3861 JOURNAL_FILE_SIZE_MIN
,
3862 m
->max_size
?: UINT64_MAX
);
3864 if (m
->keep_free
== (uint64_t) -1) {
3866 m
->keep_free
= MIN(PAGE_ALIGN(fs_size
/ 20), /* 5% of file system size */
3869 m
->keep_free
= DEFAULT_KEEP_FREE
;
3872 if (m
->n_max_files
== (uint64_t) -1)
3873 m
->n_max_files
= DEFAULT_N_MAX_FILES
;
3875 log_debug("Fixed min_use=%s max_use=%s max_size=%s min_size=%s keep_free=%s n_max_files=%" PRIu64
,
3876 format_bytes(a
, sizeof(a
), m
->min_use
),
3877 format_bytes(b
, sizeof(b
), m
->max_use
),
3878 format_bytes(c
, sizeof(c
), m
->max_size
),
3879 format_bytes(d
, sizeof(d
), m
->min_size
),
3880 format_bytes(e
, sizeof(e
), m
->keep_free
),
3884 int journal_file_get_cutoff_realtime_usec(JournalFile
*f
, usec_t
*from
, usec_t
*to
) {
3890 if (f
->header
->head_entry_realtime
== 0)
3893 *from
= le64toh(f
->header
->head_entry_realtime
);
3897 if (f
->header
->tail_entry_realtime
== 0)
3900 *to
= le64toh(f
->header
->tail_entry_realtime
);
3906 int journal_file_get_cutoff_monotonic_usec(JournalFile
*f
, sd_id128_t boot_id
, usec_t
*from
, usec_t
*to
) {
3914 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, &p
);
3918 if (le64toh(o
->data
.n_entries
) <= 0)
3922 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, le64toh(o
->data
.entry_offset
), &o
);
3926 *from
= le64toh(o
->entry
.monotonic
);
3930 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
3934 r
= generic_array_get_plus_one(f
,
3935 le64toh(o
->data
.entry_offset
),
3936 le64toh(o
->data
.entry_array_offset
),
3937 le64toh(o
->data
.n_entries
)-1,
3942 *to
= le64toh(o
->entry
.monotonic
);
3948 bool journal_file_rotate_suggested(JournalFile
*f
, usec_t max_file_usec
) {
3952 /* If we gained new header fields we gained new features,
3953 * hence suggest a rotation */
3954 if (le64toh(f
->header
->header_size
) < sizeof(Header
)) {
3955 log_debug("%s uses an outdated header, suggesting rotation.", f
->path
);
3959 /* Let's check if the hash tables grew over a certain fill
3960 * level (75%, borrowing this value from Java's hash table
3961 * implementation), and if so suggest a rotation. To calculate
3962 * the fill level we need the n_data field, which only exists
3963 * in newer versions. */
3965 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
3966 if (le64toh(f
->header
->n_data
) * 4ULL > (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
3967 log_debug("Data hash table of %s has a fill level at %.1f (%"PRIu64
" of %"PRIu64
" items, %llu file size, %"PRIu64
" bytes per hash table item), suggesting rotation.",
3969 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))),
3970 le64toh(f
->header
->n_data
),
3971 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
3972 (unsigned long long) f
->last_stat
.st_size
,
3973 f
->last_stat
.st_size
/ le64toh(f
->header
->n_data
));
3977 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
3978 if (le64toh(f
->header
->n_fields
) * 4ULL > (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
3979 log_debug("Field hash table of %s has a fill level at %.1f (%"PRIu64
" of %"PRIu64
" items), suggesting rotation.",
3981 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))),
3982 le64toh(f
->header
->n_fields
),
3983 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
));
3987 /* Are the data objects properly indexed by field objects? */
3988 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
) &&
3989 JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
) &&
3990 le64toh(f
->header
->n_data
) > 0 &&
3991 le64toh(f
->header
->n_fields
) == 0)
3994 if (max_file_usec
> 0) {
3997 h
= le64toh(f
->header
->head_entry_realtime
);
3998 t
= now(CLOCK_REALTIME
);
4000 if (h
> 0 && t
> h
+ max_file_usec
)