1 /* SPDX-License-Identifier: LGPL-2.1+ */
3 This file is part of systemd.
5 Copyright 2011 Lennart Poettering
14 #include <sys/statvfs.h>
18 #include "alloc-util.h"
19 #include "btrfs-util.h"
20 #include "chattr-util.h"
24 #include "journal-authenticate.h"
25 #include "journal-def.h"
26 #include "journal-file.h"
28 #include "parse-util.h"
29 #include "path-util.h"
30 #include "random-util.h"
33 #include "stat-util.h"
34 #include "string-util.h"
36 #include "xattr-util.h"
38 #define DEFAULT_DATA_HASH_TABLE_SIZE (2047ULL*sizeof(HashItem))
39 #define DEFAULT_FIELD_HASH_TABLE_SIZE (333ULL*sizeof(HashItem))
41 #define DEFAULT_COMPRESS_THRESHOLD (512ULL)
42 #define MIN_COMPRESS_THRESHOLD (8ULL)
44 /* This is the minimum journal file size */
45 #define JOURNAL_FILE_SIZE_MIN (512ULL*1024ULL) /* 512 KiB */
47 /* These are the lower and upper bounds if we deduce the max_use value
48 * from the file system size */
49 #define DEFAULT_MAX_USE_LOWER (1ULL*1024ULL*1024ULL) /* 1 MiB */
50 #define DEFAULT_MAX_USE_UPPER (4ULL*1024ULL*1024ULL*1024ULL) /* 4 GiB */
52 /* This is the default minimal use limit, how much we'll use even if keep_free suggests otherwise. */
53 #define DEFAULT_MIN_USE (1ULL*1024ULL*1024ULL) /* 1 MiB */
55 /* This is the upper bound if we deduce max_size from max_use */
56 #define DEFAULT_MAX_SIZE_UPPER (128ULL*1024ULL*1024ULL) /* 128 MiB */
58 /* This is the upper bound if we deduce the keep_free value from the
60 #define DEFAULT_KEEP_FREE_UPPER (4ULL*1024ULL*1024ULL*1024ULL) /* 4 GiB */
62 /* This is the keep_free value when we can't determine the system
64 #define DEFAULT_KEEP_FREE (1024ULL*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 (8ULL*1024ULL*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 if (sigfillset(&ss
) < 0)
246 r
= pthread_sigmask(SIG_BLOCK
, &ss
, &saved_ss
);
250 r
= pthread_create(&f
->offline_thread
, NULL
, journal_file_set_offline_thread
, f
);
252 k
= pthread_sigmask(SIG_SETMASK
, &saved_ss
, NULL
);
254 f
->offline_state
= OFFLINE_JOINED
;
264 static int journal_file_set_online(JournalFile
*f
) {
272 if (!(f
->fd
>= 0 && f
->header
))
276 switch (f
->offline_state
) {
278 /* No offline thread, no need to wait. */
282 case OFFLINE_SYNCING
:
283 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_SYNCING
, OFFLINE_CANCEL
))
285 /* Canceled syncing prior to offlining, no need to wait. */
288 case OFFLINE_AGAIN_FROM_SYNCING
:
289 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_AGAIN_FROM_SYNCING
, OFFLINE_CANCEL
))
291 /* Canceled restart from syncing, no need to wait. */
294 case OFFLINE_AGAIN_FROM_OFFLINING
:
295 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_AGAIN_FROM_OFFLINING
, OFFLINE_CANCEL
))
297 /* Canceled restart from offlining, must wait for offlining to complete however. */
302 r
= journal_file_set_offline_thread_join(f
);
312 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
315 switch (f
->header
->state
) {
320 f
->header
->state
= STATE_ONLINE
;
329 bool journal_file_is_offlining(JournalFile
*f
) {
332 __sync_synchronize();
334 if (IN_SET(f
->offline_state
, OFFLINE_DONE
, OFFLINE_JOINED
))
340 JournalFile
* journal_file_close(JournalFile
*f
) {
344 /* Write the final tag */
345 if (f
->seal
&& f
->writable
) {
348 r
= journal_file_append_tag(f
);
350 log_error_errno(r
, "Failed to append tag when closing journal: %m");
354 if (f
->post_change_timer
) {
357 if (sd_event_source_get_enabled(f
->post_change_timer
, &enabled
) >= 0)
358 if (enabled
== SD_EVENT_ONESHOT
)
359 journal_file_post_change(f
);
361 (void) sd_event_source_set_enabled(f
->post_change_timer
, SD_EVENT_OFF
);
362 sd_event_source_unref(f
->post_change_timer
);
365 journal_file_set_offline(f
, true);
367 if (f
->mmap
&& f
->cache_fd
)
368 mmap_cache_free_fd(f
->mmap
, f
->cache_fd
);
370 if (f
->fd
>= 0 && f
->defrag_on_close
) {
372 /* Be friendly to btrfs: turn COW back on again now,
373 * and defragment the file. We won't write to the file
374 * ever again, hence remove all fragmentation, and
375 * reenable all the good bits COW usually provides
376 * (such as data checksumming). */
378 (void) chattr_fd(f
->fd
, 0, FS_NOCOW_FL
);
379 (void) btrfs_defrag_fd(f
->fd
);
386 mmap_cache_unref(f
->mmap
);
388 ordered_hashmap_free_free(f
->chain_cache
);
390 #if HAVE_XZ || HAVE_LZ4
391 free(f
->compress_buffer
);
396 munmap(f
->fss_file
, PAGE_ALIGN(f
->fss_file_size
));
398 free(f
->fsprg_state
);
403 gcry_md_close(f
->hmac
);
409 static int journal_file_init_header(JournalFile
*f
, JournalFile
*template) {
416 memcpy(h
.signature
, HEADER_SIGNATURE
, 8);
417 h
.header_size
= htole64(ALIGN64(sizeof(h
)));
419 h
.incompatible_flags
|= htole32(
420 f
->compress_xz
* HEADER_INCOMPATIBLE_COMPRESSED_XZ
|
421 f
->compress_lz4
* HEADER_INCOMPATIBLE_COMPRESSED_LZ4
);
423 h
.compatible_flags
= htole32(
424 f
->seal
* HEADER_COMPATIBLE_SEALED
);
426 r
= sd_id128_randomize(&h
.file_id
);
431 h
.seqnum_id
= template->header
->seqnum_id
;
432 h
.tail_entry_seqnum
= template->header
->tail_entry_seqnum
;
434 h
.seqnum_id
= h
.file_id
;
436 k
= pwrite(f
->fd
, &h
, sizeof(h
), 0);
446 static int journal_file_refresh_header(JournalFile
*f
) {
453 r
= sd_id128_get_machine(&f
->header
->machine_id
);
457 r
= sd_id128_get_boot(&boot_id
);
461 f
->header
->boot_id
= boot_id
;
463 r
= journal_file_set_online(f
);
465 /* Sync the online state to disk */
468 /* We likely just created a new file, also sync the directory this file is located in. */
469 (void) fsync_directory_of_file(f
->fd
);
474 static bool warn_wrong_flags(const JournalFile
*f
, bool compatible
) {
475 const uint32_t any
= compatible
? HEADER_COMPATIBLE_ANY
: HEADER_INCOMPATIBLE_ANY
,
476 supported
= compatible
? HEADER_COMPATIBLE_SUPPORTED
: HEADER_INCOMPATIBLE_SUPPORTED
;
477 const char *type
= compatible
? "compatible" : "incompatible";
480 flags
= le32toh(compatible
? f
->header
->compatible_flags
: f
->header
->incompatible_flags
);
482 if (flags
& ~supported
) {
484 log_debug("Journal file %s has unknown %s flags 0x%"PRIx32
,
485 f
->path
, type
, flags
& ~any
);
486 flags
= (flags
& any
) & ~supported
;
490 _cleanup_free_
char *t
= NULL
;
492 if (compatible
&& (flags
& HEADER_COMPATIBLE_SEALED
))
493 strv
[n
++] = "sealed";
494 if (!compatible
&& (flags
& HEADER_INCOMPATIBLE_COMPRESSED_XZ
))
495 strv
[n
++] = "xz-compressed";
496 if (!compatible
&& (flags
& HEADER_INCOMPATIBLE_COMPRESSED_LZ4
))
497 strv
[n
++] = "lz4-compressed";
499 assert(n
< ELEMENTSOF(strv
));
501 t
= strv_join((char**) strv
, ", ");
502 log_debug("Journal file %s uses %s %s %s disabled at compilation time.",
503 f
->path
, type
, n
> 1 ? "flags" : "flag", strnull(t
));
511 static int journal_file_verify_header(JournalFile
*f
) {
512 uint64_t arena_size
, header_size
;
517 if (memcmp(f
->header
->signature
, HEADER_SIGNATURE
, 8))
520 /* In both read and write mode we refuse to open files with incompatible
521 * flags we don't know. */
522 if (warn_wrong_flags(f
, false))
523 return -EPROTONOSUPPORT
;
525 /* When open for writing we refuse to open files with compatible flags, too. */
526 if (f
->writable
&& warn_wrong_flags(f
, true))
527 return -EPROTONOSUPPORT
;
529 if (f
->header
->state
>= _STATE_MAX
)
532 header_size
= le64toh(f
->header
->header_size
);
534 /* The first addition was n_data, so check that we are at least this large */
535 if (header_size
< HEADER_SIZE_MIN
)
538 if (JOURNAL_HEADER_SEALED(f
->header
) && !JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
541 arena_size
= le64toh(f
->header
->arena_size
);
543 if (UINT64_MAX
- header_size
< arena_size
|| header_size
+ arena_size
> (uint64_t) f
->last_stat
.st_size
)
546 if (le64toh(f
->header
->tail_object_offset
) > header_size
+ arena_size
)
549 if (!VALID64(le64toh(f
->header
->data_hash_table_offset
)) ||
550 !VALID64(le64toh(f
->header
->field_hash_table_offset
)) ||
551 !VALID64(le64toh(f
->header
->tail_object_offset
)) ||
552 !VALID64(le64toh(f
->header
->entry_array_offset
)))
556 sd_id128_t machine_id
;
560 r
= sd_id128_get_machine(&machine_id
);
564 if (!sd_id128_equal(machine_id
, f
->header
->machine_id
))
567 state
= f
->header
->state
;
569 if (state
== STATE_ARCHIVED
)
570 return -ESHUTDOWN
; /* Already archived */
571 else if (state
== STATE_ONLINE
) {
572 log_debug("Journal file %s is already online. Assuming unclean closing.", f
->path
);
574 } else if (state
!= STATE_OFFLINE
) {
575 log_debug("Journal file %s has unknown state %i.", f
->path
, state
);
579 if (f
->header
->field_hash_table_size
== 0 || f
->header
->data_hash_table_size
== 0)
582 /* Don't permit appending to files from the future. Because otherwise the realtime timestamps wouldn't
583 * be strictly ordered in the entries in the file anymore, and we can't have that since it breaks
585 if (le64toh(f
->header
->tail_entry_realtime
) > now(CLOCK_REALTIME
)) {
586 log_debug("Journal file %s is from the future, refusing to append new data to it that'd be older.", f
->path
);
591 f
->compress_xz
= JOURNAL_HEADER_COMPRESSED_XZ(f
->header
);
592 f
->compress_lz4
= JOURNAL_HEADER_COMPRESSED_LZ4(f
->header
);
594 f
->seal
= JOURNAL_HEADER_SEALED(f
->header
);
599 static int journal_file_fstat(JournalFile
*f
) {
605 if (fstat(f
->fd
, &f
->last_stat
) < 0)
608 f
->last_stat_usec
= now(CLOCK_MONOTONIC
);
610 /* Refuse dealing with with files that aren't regular */
611 r
= stat_verify_regular(&f
->last_stat
);
615 /* Refuse appending to files that are already deleted */
616 if (f
->last_stat
.st_nlink
<= 0)
622 static int journal_file_allocate(JournalFile
*f
, uint64_t offset
, uint64_t size
) {
623 uint64_t old_size
, new_size
;
629 /* We assume that this file is not sparse, and we know that
630 * for sure, since we always call posix_fallocate()
633 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
637 le64toh(f
->header
->header_size
) +
638 le64toh(f
->header
->arena_size
);
640 new_size
= PAGE_ALIGN(offset
+ size
);
641 if (new_size
< le64toh(f
->header
->header_size
))
642 new_size
= le64toh(f
->header
->header_size
);
644 if (new_size
<= old_size
) {
646 /* We already pre-allocated enough space, but before
647 * we write to it, let's check with fstat() if the
648 * file got deleted, in order make sure we don't throw
649 * away the data immediately. Don't check fstat() for
650 * all writes though, but only once ever 10s. */
652 if (f
->last_stat_usec
+ LAST_STAT_REFRESH_USEC
> now(CLOCK_MONOTONIC
))
655 return journal_file_fstat(f
);
658 /* Allocate more space. */
660 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
663 if (new_size
> f
->metrics
.min_size
&& f
->metrics
.keep_free
> 0) {
666 if (fstatvfs(f
->fd
, &svfs
) >= 0) {
669 available
= LESS_BY((uint64_t) svfs
.f_bfree
* (uint64_t) svfs
.f_bsize
, f
->metrics
.keep_free
);
671 if (new_size
- old_size
> available
)
676 /* Increase by larger blocks at once */
677 new_size
= DIV_ROUND_UP(new_size
, FILE_SIZE_INCREASE
) * FILE_SIZE_INCREASE
;
678 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
679 new_size
= f
->metrics
.max_size
;
681 /* Note that the glibc fallocate() fallback is very
682 inefficient, hence we try to minimize the allocation area
684 r
= posix_fallocate(f
->fd
, old_size
, new_size
- old_size
);
688 f
->header
->arena_size
= htole64(new_size
- le64toh(f
->header
->header_size
));
690 return journal_file_fstat(f
);
693 static unsigned type_to_context(ObjectType type
) {
694 /* One context for each type, plus one catch-all for the rest */
695 assert_cc(_OBJECT_TYPE_MAX
<= MMAP_CACHE_MAX_CONTEXTS
);
696 assert_cc(CONTEXT_HEADER
< MMAP_CACHE_MAX_CONTEXTS
);
697 return type
> OBJECT_UNUSED
&& type
< _OBJECT_TYPE_MAX
? type
: 0;
700 static int journal_file_move_to(JournalFile
*f
, ObjectType type
, bool keep_always
, uint64_t offset
, uint64_t size
, void **ret
, size_t *ret_size
) {
709 /* Avoid SIGBUS on invalid accesses */
710 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
) {
711 /* Hmm, out of range? Let's refresh the fstat() data
712 * first, before we trust that check. */
714 r
= journal_file_fstat(f
);
718 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
)
719 return -EADDRNOTAVAIL
;
722 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
);
725 static uint64_t minimum_header_size(Object
*o
) {
727 static const uint64_t table
[] = {
728 [OBJECT_DATA
] = sizeof(DataObject
),
729 [OBJECT_FIELD
] = sizeof(FieldObject
),
730 [OBJECT_ENTRY
] = sizeof(EntryObject
),
731 [OBJECT_DATA_HASH_TABLE
] = sizeof(HashTableObject
),
732 [OBJECT_FIELD_HASH_TABLE
] = sizeof(HashTableObject
),
733 [OBJECT_ENTRY_ARRAY
] = sizeof(EntryArrayObject
),
734 [OBJECT_TAG
] = sizeof(TagObject
),
737 if (o
->object
.type
>= ELEMENTSOF(table
) || table
[o
->object
.type
] <= 0)
738 return sizeof(ObjectHeader
);
740 return table
[o
->object
.type
];
743 /* Lightweight object checks. We want this to be fast, so that we won't
744 * slowdown every journal_file_move_to_object() call too much. */
745 static int journal_file_check_object(JournalFile
*f
, uint64_t offset
, Object
*o
) {
749 switch (o
->object
.type
) {
752 if ((le64toh(o
->data
.entry_offset
) == 0) ^ (le64toh(o
->data
.n_entries
) == 0)) {
753 log_debug("Bad n_entries: %"PRIu64
": %"PRIu64
,
754 le64toh(o
->data
.n_entries
), offset
);
758 if (le64toh(o
->object
.size
) - offsetof(DataObject
, payload
) <= 0) {
759 log_debug("Bad object size (<= %zu): %"PRIu64
": %"PRIu64
,
760 offsetof(DataObject
, payload
),
761 le64toh(o
->object
.size
),
766 if (!VALID64(le64toh(o
->data
.next_hash_offset
)) ||
767 !VALID64(le64toh(o
->data
.next_field_offset
)) ||
768 !VALID64(le64toh(o
->data
.entry_offset
)) ||
769 !VALID64(le64toh(o
->data
.entry_array_offset
))) {
770 log_debug("Invalid offset, next_hash_offset="OFSfmt
", next_field_offset="OFSfmt
771 ", entry_offset="OFSfmt
", entry_array_offset="OFSfmt
": %"PRIu64
,
772 le64toh(o
->data
.next_hash_offset
),
773 le64toh(o
->data
.next_field_offset
),
774 le64toh(o
->data
.entry_offset
),
775 le64toh(o
->data
.entry_array_offset
),
784 if (le64toh(o
->object
.size
) - offsetof(FieldObject
, payload
) <= 0) {
786 "Bad field size (<= %zu): %"PRIu64
": %"PRIu64
,
787 offsetof(FieldObject
, payload
),
788 le64toh(o
->object
.size
),
793 if (!VALID64(le64toh(o
->field
.next_hash_offset
)) ||
794 !VALID64(le64toh(o
->field
.head_data_offset
))) {
796 "Invalid offset, next_hash_offset="OFSfmt
797 ", head_data_offset="OFSfmt
": %"PRIu64
,
798 le64toh(o
->field
.next_hash_offset
),
799 le64toh(o
->field
.head_data_offset
),
806 if ((le64toh(o
->object
.size
) - offsetof(EntryObject
, items
)) % sizeof(EntryItem
) != 0) {
808 "Bad entry size (<= %zu): %"PRIu64
": %"PRIu64
,
809 offsetof(EntryObject
, items
),
810 le64toh(o
->object
.size
),
815 if ((le64toh(o
->object
.size
) - offsetof(EntryObject
, items
)) / sizeof(EntryItem
) <= 0) {
817 "Invalid number items in entry: %"PRIu64
": %"PRIu64
,
818 (le64toh(o
->object
.size
) - offsetof(EntryObject
, items
)) / sizeof(EntryItem
),
823 if (le64toh(o
->entry
.seqnum
) <= 0) {
825 "Invalid entry seqnum: %"PRIx64
": %"PRIu64
,
826 le64toh(o
->entry
.seqnum
),
831 if (!VALID_REALTIME(le64toh(o
->entry
.realtime
))) {
833 "Invalid entry realtime timestamp: %"PRIu64
": %"PRIu64
,
834 le64toh(o
->entry
.realtime
),
839 if (!VALID_MONOTONIC(le64toh(o
->entry
.monotonic
))) {
841 "Invalid entry monotonic timestamp: %"PRIu64
": %"PRIu64
,
842 le64toh(o
->entry
.monotonic
),
849 case OBJECT_DATA_HASH_TABLE
:
850 case OBJECT_FIELD_HASH_TABLE
:
851 if ((le64toh(o
->object
.size
) - offsetof(HashTableObject
, items
)) % sizeof(HashItem
) != 0 ||
852 (le64toh(o
->object
.size
) - offsetof(HashTableObject
, items
)) / sizeof(HashItem
) <= 0) {
854 "Invalid %s hash table size: %"PRIu64
": %"PRIu64
,
855 o
->object
.type
== OBJECT_DATA_HASH_TABLE
? "data" : "field",
856 le64toh(o
->object
.size
),
863 case OBJECT_ENTRY_ARRAY
:
864 if ((le64toh(o
->object
.size
) - offsetof(EntryArrayObject
, items
)) % sizeof(le64_t
) != 0 ||
865 (le64toh(o
->object
.size
) - offsetof(EntryArrayObject
, items
)) / sizeof(le64_t
) <= 0) {
867 "Invalid object entry array size: %"PRIu64
": %"PRIu64
,
868 le64toh(o
->object
.size
),
873 if (!VALID64(le64toh(o
->entry_array
.next_entry_array_offset
))) {
875 "Invalid object entry array next_entry_array_offset: "OFSfmt
": %"PRIu64
,
876 le64toh(o
->entry_array
.next_entry_array_offset
),
884 if (le64toh(o
->object
.size
) != sizeof(TagObject
)) {
886 "Invalid object tag size: %"PRIu64
": %"PRIu64
,
887 le64toh(o
->object
.size
),
892 if (!VALID_EPOCH(le64toh(o
->tag
.epoch
))) {
894 "Invalid object tag epoch: %"PRIu64
": %"PRIu64
,
895 le64toh(o
->tag
.epoch
),
906 int journal_file_move_to_object(JournalFile
*f
, ObjectType type
, uint64_t offset
, Object
**ret
) {
916 /* Objects may only be located at multiple of 64 bit */
917 if (!VALID64(offset
)) {
918 log_debug("Attempt to move to object at non-64bit boundary: %" PRIu64
, offset
);
922 /* Object may not be located in the file header */
923 if (offset
< le64toh(f
->header
->header_size
)) {
924 log_debug("Attempt to move to object located in file header: %" PRIu64
, offset
);
928 r
= journal_file_move_to(f
, type
, false, offset
, sizeof(ObjectHeader
), &t
, &tsize
);
933 s
= le64toh(o
->object
.size
);
936 log_debug("Attempt to move to uninitialized object: %" PRIu64
, offset
);
939 if (s
< sizeof(ObjectHeader
)) {
940 log_debug("Attempt to move to overly short object: %" PRIu64
, offset
);
944 if (o
->object
.type
<= OBJECT_UNUSED
) {
945 log_debug("Attempt to move to object with invalid type: %" PRIu64
, offset
);
949 if (s
< minimum_header_size(o
)) {
950 log_debug("Attempt to move to truncated object: %" PRIu64
, offset
);
954 if (type
> OBJECT_UNUSED
&& o
->object
.type
!= type
) {
955 log_debug("Attempt to move to object of unexpected type: %" PRIu64
, offset
);
960 r
= journal_file_move_to(f
, type
, false, offset
, s
, &t
, NULL
);
967 r
= journal_file_check_object(f
, offset
, o
);
975 static uint64_t journal_file_entry_seqnum(JournalFile
*f
, uint64_t *seqnum
) {
981 r
= le64toh(f
->header
->tail_entry_seqnum
) + 1;
984 /* If an external seqnum counter was passed, we update
985 * both the local and the external one, and set it to
986 * the maximum of both */
994 f
->header
->tail_entry_seqnum
= htole64(r
);
996 if (f
->header
->head_entry_seqnum
== 0)
997 f
->header
->head_entry_seqnum
= htole64(r
);
1002 int journal_file_append_object(JournalFile
*f
, ObjectType type
, uint64_t size
, Object
**ret
, uint64_t *offset
) {
1010 assert(type
> OBJECT_UNUSED
&& type
< _OBJECT_TYPE_MAX
);
1011 assert(size
>= sizeof(ObjectHeader
));
1015 r
= journal_file_set_online(f
);
1019 p
= le64toh(f
->header
->tail_object_offset
);
1021 p
= le64toh(f
->header
->header_size
);
1023 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &tail
);
1027 p
+= ALIGN64(le64toh(tail
->object
.size
));
1030 r
= journal_file_allocate(f
, p
, size
);
1034 r
= journal_file_move_to(f
, type
, false, p
, size
, &t
, NULL
);
1041 o
->object
.type
= type
;
1042 o
->object
.size
= htole64(size
);
1044 f
->header
->tail_object_offset
= htole64(p
);
1045 f
->header
->n_objects
= htole64(le64toh(f
->header
->n_objects
) + 1);
1053 static int journal_file_setup_data_hash_table(JournalFile
*f
) {
1061 /* We estimate that we need 1 hash table entry per 768 bytes
1062 of journal file and we want to make sure we never get
1063 beyond 75% fill level. Calculate the hash table size for
1064 the maximum file size based on these metrics. */
1066 s
= (f
->metrics
.max_size
* 4 / 768 / 3) * sizeof(HashItem
);
1067 if (s
< DEFAULT_DATA_HASH_TABLE_SIZE
)
1068 s
= DEFAULT_DATA_HASH_TABLE_SIZE
;
1070 log_debug("Reserving %"PRIu64
" entries in hash table.", s
/ sizeof(HashItem
));
1072 r
= journal_file_append_object(f
,
1073 OBJECT_DATA_HASH_TABLE
,
1074 offsetof(Object
, hash_table
.items
) + s
,
1079 memzero(o
->hash_table
.items
, s
);
1081 f
->header
->data_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
1082 f
->header
->data_hash_table_size
= htole64(s
);
1087 static int journal_file_setup_field_hash_table(JournalFile
*f
) {
1095 /* We use a fixed size hash table for the fields as this
1096 * number should grow very slowly only */
1098 s
= DEFAULT_FIELD_HASH_TABLE_SIZE
;
1099 r
= journal_file_append_object(f
,
1100 OBJECT_FIELD_HASH_TABLE
,
1101 offsetof(Object
, hash_table
.items
) + s
,
1106 memzero(o
->hash_table
.items
, s
);
1108 f
->header
->field_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
1109 f
->header
->field_hash_table_size
= htole64(s
);
1114 int journal_file_map_data_hash_table(JournalFile
*f
) {
1122 if (f
->data_hash_table
)
1125 p
= le64toh(f
->header
->data_hash_table_offset
);
1126 s
= le64toh(f
->header
->data_hash_table_size
);
1128 r
= journal_file_move_to(f
,
1129 OBJECT_DATA_HASH_TABLE
,
1136 f
->data_hash_table
= t
;
1140 int journal_file_map_field_hash_table(JournalFile
*f
) {
1148 if (f
->field_hash_table
)
1151 p
= le64toh(f
->header
->field_hash_table_offset
);
1152 s
= le64toh(f
->header
->field_hash_table_size
);
1154 r
= journal_file_move_to(f
,
1155 OBJECT_FIELD_HASH_TABLE
,
1162 f
->field_hash_table
= t
;
1166 static int journal_file_link_field(
1177 assert(f
->field_hash_table
);
1181 if (o
->object
.type
!= OBJECT_FIELD
)
1184 m
= le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
);
1188 /* This might alter the window we are looking at */
1189 o
->field
.next_hash_offset
= o
->field
.head_data_offset
= 0;
1192 p
= le64toh(f
->field_hash_table
[h
].tail_hash_offset
);
1194 f
->field_hash_table
[h
].head_hash_offset
= htole64(offset
);
1196 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1200 o
->field
.next_hash_offset
= htole64(offset
);
1203 f
->field_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1205 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
1206 f
->header
->n_fields
= htole64(le64toh(f
->header
->n_fields
) + 1);
1211 static int journal_file_link_data(
1222 assert(f
->data_hash_table
);
1226 if (o
->object
.type
!= OBJECT_DATA
)
1229 m
= le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
);
1233 /* This might alter the window we are looking at */
1234 o
->data
.next_hash_offset
= o
->data
.next_field_offset
= 0;
1235 o
->data
.entry_offset
= o
->data
.entry_array_offset
= 0;
1236 o
->data
.n_entries
= 0;
1239 p
= le64toh(f
->data_hash_table
[h
].tail_hash_offset
);
1241 /* Only entry in the hash table is easy */
1242 f
->data_hash_table
[h
].head_hash_offset
= htole64(offset
);
1244 /* Move back to the previous data object, to patch in
1247 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1251 o
->data
.next_hash_offset
= htole64(offset
);
1254 f
->data_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1256 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
1257 f
->header
->n_data
= htole64(le64toh(f
->header
->n_data
) + 1);
1262 int journal_file_find_field_object_with_hash(
1264 const void *field
, uint64_t size
, uint64_t hash
,
1265 Object
**ret
, uint64_t *offset
) {
1267 uint64_t p
, osize
, h
, m
;
1272 assert(field
&& size
> 0);
1274 /* If the field hash table is empty, we can't find anything */
1275 if (le64toh(f
->header
->field_hash_table_size
) <= 0)
1278 /* Map the field hash table, if it isn't mapped yet. */
1279 r
= journal_file_map_field_hash_table(f
);
1283 osize
= offsetof(Object
, field
.payload
) + size
;
1285 m
= le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
);
1290 p
= le64toh(f
->field_hash_table
[h
].head_hash_offset
);
1295 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1299 if (le64toh(o
->field
.hash
) == hash
&&
1300 le64toh(o
->object
.size
) == osize
&&
1301 memcmp(o
->field
.payload
, field
, size
) == 0) {
1311 p
= le64toh(o
->field
.next_hash_offset
);
1317 int journal_file_find_field_object(
1319 const void *field
, uint64_t size
,
1320 Object
**ret
, uint64_t *offset
) {
1325 assert(field
&& size
> 0);
1327 hash
= hash64(field
, size
);
1329 return journal_file_find_field_object_with_hash(f
,
1334 int journal_file_find_data_object_with_hash(
1336 const void *data
, uint64_t size
, uint64_t hash
,
1337 Object
**ret
, uint64_t *offset
) {
1339 uint64_t p
, osize
, h
, m
;
1344 assert(data
|| size
== 0);
1346 /* If there's no data hash table, then there's no entry. */
1347 if (le64toh(f
->header
->data_hash_table_size
) <= 0)
1350 /* Map the data hash table, if it isn't mapped yet. */
1351 r
= journal_file_map_data_hash_table(f
);
1355 osize
= offsetof(Object
, data
.payload
) + size
;
1357 m
= le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
);
1362 p
= le64toh(f
->data_hash_table
[h
].head_hash_offset
);
1367 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1371 if (le64toh(o
->data
.hash
) != hash
)
1374 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
) {
1375 #if HAVE_XZ || HAVE_LZ4
1379 l
= le64toh(o
->object
.size
);
1380 if (l
<= offsetof(Object
, data
.payload
))
1383 l
-= offsetof(Object
, data
.payload
);
1385 r
= decompress_blob(o
->object
.flags
& OBJECT_COMPRESSION_MASK
,
1386 o
->data
.payload
, l
, &f
->compress_buffer
, &f
->compress_buffer_size
, &rsize
, 0);
1390 if (rsize
== size
&&
1391 memcmp(f
->compress_buffer
, data
, size
) == 0) {
1402 return -EPROTONOSUPPORT
;
1404 } else if (le64toh(o
->object
.size
) == osize
&&
1405 memcmp(o
->data
.payload
, data
, size
) == 0) {
1417 p
= le64toh(o
->data
.next_hash_offset
);
1423 int journal_file_find_data_object(
1425 const void *data
, uint64_t size
,
1426 Object
**ret
, uint64_t *offset
) {
1431 assert(data
|| size
== 0);
1433 hash
= hash64(data
, size
);
1435 return journal_file_find_data_object_with_hash(f
,
1440 static int journal_file_append_field(
1442 const void *field
, uint64_t size
,
1443 Object
**ret
, uint64_t *offset
) {
1451 assert(field
&& size
> 0);
1453 hash
= hash64(field
, size
);
1455 r
= journal_file_find_field_object_with_hash(f
, field
, size
, hash
, &o
, &p
);
1469 osize
= offsetof(Object
, field
.payload
) + size
;
1470 r
= journal_file_append_object(f
, OBJECT_FIELD
, osize
, &o
, &p
);
1474 o
->field
.hash
= htole64(hash
);
1475 memcpy(o
->field
.payload
, field
, size
);
1477 r
= journal_file_link_field(f
, o
, p
, hash
);
1481 /* The linking might have altered the window, so let's
1482 * refresh our pointer */
1483 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1488 r
= journal_file_hmac_put_object(f
, OBJECT_FIELD
, o
, p
);
1502 static int journal_file_append_data(
1504 const void *data
, uint64_t size
,
1505 Object
**ret
, uint64_t *offset
) {
1510 int r
, compression
= 0;
1514 assert(data
|| size
== 0);
1516 hash
= hash64(data
, size
);
1518 r
= journal_file_find_data_object_with_hash(f
, data
, size
, hash
, &o
, &p
);
1532 osize
= offsetof(Object
, data
.payload
) + size
;
1533 r
= journal_file_append_object(f
, OBJECT_DATA
, osize
, &o
, &p
);
1537 o
->data
.hash
= htole64(hash
);
1539 #if HAVE_XZ || HAVE_LZ4
1540 if (JOURNAL_FILE_COMPRESS(f
) && size
>= f
->compress_threshold_bytes
) {
1543 compression
= compress_blob(data
, size
, o
->data
.payload
, size
- 1, &rsize
);
1545 if (compression
>= 0) {
1546 o
->object
.size
= htole64(offsetof(Object
, data
.payload
) + rsize
);
1547 o
->object
.flags
|= compression
;
1549 log_debug("Compressed data object %"PRIu64
" -> %zu using %s",
1550 size
, rsize
, object_compressed_to_string(compression
));
1552 /* Compression didn't work, we don't really care why, let's continue without compression */
1557 if (compression
== 0)
1558 memcpy_safe(o
->data
.payload
, data
, size
);
1560 r
= journal_file_link_data(f
, o
, p
, hash
);
1565 r
= journal_file_hmac_put_object(f
, OBJECT_DATA
, o
, p
);
1570 /* The linking might have altered the window, so let's
1571 * refresh our pointer */
1572 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1579 eq
= memchr(data
, '=', size
);
1580 if (eq
&& eq
> data
) {
1584 /* Create field object ... */
1585 r
= journal_file_append_field(f
, data
, (uint8_t*) eq
- (uint8_t*) data
, &fo
, &fp
);
1589 /* ... and link it in. */
1590 o
->data
.next_field_offset
= fo
->field
.head_data_offset
;
1591 fo
->field
.head_data_offset
= le64toh(p
);
1603 uint64_t journal_file_entry_n_items(Object
*o
) {
1606 if (o
->object
.type
!= OBJECT_ENTRY
)
1609 return (le64toh(o
->object
.size
) - offsetof(Object
, entry
.items
)) / sizeof(EntryItem
);
1612 uint64_t journal_file_entry_array_n_items(Object
*o
) {
1615 if (o
->object
.type
!= OBJECT_ENTRY_ARRAY
)
1618 return (le64toh(o
->object
.size
) - offsetof(Object
, entry_array
.items
)) / sizeof(uint64_t);
1621 uint64_t journal_file_hash_table_n_items(Object
*o
) {
1624 if (!IN_SET(o
->object
.type
, OBJECT_DATA_HASH_TABLE
, OBJECT_FIELD_HASH_TABLE
))
1627 return (le64toh(o
->object
.size
) - offsetof(Object
, hash_table
.items
)) / sizeof(HashItem
);
1630 static int link_entry_into_array(JournalFile
*f
,
1635 uint64_t n
= 0, ap
= 0, q
, i
, a
, hidx
;
1644 a
= le64toh(*first
);
1645 i
= hidx
= le64toh(*idx
);
1648 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
1652 n
= journal_file_entry_array_n_items(o
);
1654 o
->entry_array
.items
[i
] = htole64(p
);
1655 *idx
= htole64(hidx
+ 1);
1661 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
1672 r
= journal_file_append_object(f
, OBJECT_ENTRY_ARRAY
,
1673 offsetof(Object
, entry_array
.items
) + n
* sizeof(uint64_t),
1679 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY_ARRAY
, o
, q
);
1684 o
->entry_array
.items
[i
] = htole64(p
);
1687 *first
= htole64(q
);
1689 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, ap
, &o
);
1693 o
->entry_array
.next_entry_array_offset
= htole64(q
);
1696 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
1697 f
->header
->n_entry_arrays
= htole64(le64toh(f
->header
->n_entry_arrays
) + 1);
1699 *idx
= htole64(hidx
+ 1);
1704 static int link_entry_into_array_plus_one(JournalFile
*f
,
1719 *extra
= htole64(p
);
1723 i
= htole64(le64toh(*idx
) - 1);
1724 r
= link_entry_into_array(f
, first
, &i
, p
);
1729 *idx
= htole64(le64toh(*idx
) + 1);
1733 static int journal_file_link_entry_item(JournalFile
*f
, Object
*o
, uint64_t offset
, uint64_t i
) {
1740 p
= le64toh(o
->entry
.items
[i
].object_offset
);
1744 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1748 return link_entry_into_array_plus_one(f
,
1749 &o
->data
.entry_offset
,
1750 &o
->data
.entry_array_offset
,
1755 static int journal_file_link_entry(JournalFile
*f
, Object
*o
, uint64_t offset
) {
1764 if (o
->object
.type
!= OBJECT_ENTRY
)
1767 __sync_synchronize();
1769 /* Link up the entry itself */
1770 r
= link_entry_into_array(f
,
1771 &f
->header
->entry_array_offset
,
1772 &f
->header
->n_entries
,
1777 /* log_debug("=> %s seqnr=%"PRIu64" n_entries=%"PRIu64, f->path, o->entry.seqnum, f->header->n_entries); */
1779 if (f
->header
->head_entry_realtime
== 0)
1780 f
->header
->head_entry_realtime
= o
->entry
.realtime
;
1782 f
->header
->tail_entry_realtime
= o
->entry
.realtime
;
1783 f
->header
->tail_entry_monotonic
= o
->entry
.monotonic
;
1785 /* Link up the items */
1786 n
= journal_file_entry_n_items(o
);
1787 for (i
= 0; i
< n
; i
++) {
1788 r
= journal_file_link_entry_item(f
, o
, offset
, i
);
1796 static int journal_file_append_entry_internal(
1798 const dual_timestamp
*ts
,
1800 const EntryItem items
[], unsigned n_items
,
1802 Object
**ret
, uint64_t *offset
) {
1810 assert(items
|| n_items
== 0);
1813 osize
= offsetof(Object
, entry
.items
) + (n_items
* sizeof(EntryItem
));
1815 r
= journal_file_append_object(f
, OBJECT_ENTRY
, osize
, &o
, &np
);
1819 o
->entry
.seqnum
= htole64(journal_file_entry_seqnum(f
, seqnum
));
1820 memcpy_safe(o
->entry
.items
, items
, n_items
* sizeof(EntryItem
));
1821 o
->entry
.realtime
= htole64(ts
->realtime
);
1822 o
->entry
.monotonic
= htole64(ts
->monotonic
);
1823 o
->entry
.xor_hash
= htole64(xor_hash
);
1824 o
->entry
.boot_id
= f
->header
->boot_id
;
1827 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY
, o
, np
);
1832 r
= journal_file_link_entry(f
, o
, np
);
1845 void journal_file_post_change(JournalFile
*f
) {
1848 /* inotify() does not receive IN_MODIFY events from file
1849 * accesses done via mmap(). After each access we hence
1850 * trigger IN_MODIFY by truncating the journal file to its
1851 * current size which triggers IN_MODIFY. */
1853 __sync_synchronize();
1855 if (ftruncate(f
->fd
, f
->last_stat
.st_size
) < 0)
1856 log_debug_errno(errno
, "Failed to truncate file to its own size: %m");
1859 static int post_change_thunk(sd_event_source
*timer
, uint64_t usec
, void *userdata
) {
1862 journal_file_post_change(userdata
);
1867 static void schedule_post_change(JournalFile
*f
) {
1868 sd_event_source
*timer
;
1873 assert(f
->post_change_timer
);
1875 timer
= f
->post_change_timer
;
1877 r
= sd_event_source_get_enabled(timer
, &enabled
);
1879 log_debug_errno(r
, "Failed to get ftruncate timer state: %m");
1883 if (enabled
== SD_EVENT_ONESHOT
)
1886 r
= sd_event_now(sd_event_source_get_event(timer
), CLOCK_MONOTONIC
, &now
);
1888 log_debug_errno(r
, "Failed to get clock's now for scheduling ftruncate: %m");
1892 r
= sd_event_source_set_time(timer
, now
+f
->post_change_timer_period
);
1894 log_debug_errno(r
, "Failed to set time for scheduling ftruncate: %m");
1898 r
= sd_event_source_set_enabled(timer
, SD_EVENT_ONESHOT
);
1900 log_debug_errno(r
, "Failed to enable scheduled ftruncate: %m");
1907 /* On failure, let's simply post the change immediately. */
1908 journal_file_post_change(f
);
1911 /* Enable coalesced change posting in a timer on the provided sd_event instance */
1912 int journal_file_enable_post_change_timer(JournalFile
*f
, sd_event
*e
, usec_t t
) {
1913 _cleanup_(sd_event_source_unrefp
) sd_event_source
*timer
= NULL
;
1917 assert_return(!f
->post_change_timer
, -EINVAL
);
1921 r
= sd_event_add_time(e
, &timer
, CLOCK_MONOTONIC
, 0, 0, post_change_thunk
, f
);
1925 r
= sd_event_source_set_enabled(timer
, SD_EVENT_OFF
);
1929 f
->post_change_timer
= TAKE_PTR(timer
);
1930 f
->post_change_timer_period
= t
;
1935 static int entry_item_cmp(const void *_a
, const void *_b
) {
1936 const EntryItem
*a
= _a
, *b
= _b
;
1938 if (le64toh(a
->object_offset
) < le64toh(b
->object_offset
))
1940 if (le64toh(a
->object_offset
) > le64toh(b
->object_offset
))
1945 int journal_file_append_entry(JournalFile
*f
, const dual_timestamp
*ts
, const struct iovec iovec
[], unsigned n_iovec
, uint64_t *seqnum
, Object
**ret
, uint64_t *offset
) {
1949 uint64_t xor_hash
= 0;
1950 struct dual_timestamp _ts
;
1954 assert(iovec
|| n_iovec
== 0);
1957 dual_timestamp_get(&_ts
);
1962 r
= journal_file_maybe_append_tag(f
, ts
->realtime
);
1967 /* alloca() can't take 0, hence let's allocate at least one */
1968 items
= newa(EntryItem
, MAX(1u, n_iovec
));
1970 for (i
= 0; i
< n_iovec
; i
++) {
1974 r
= journal_file_append_data(f
, iovec
[i
].iov_base
, iovec
[i
].iov_len
, &o
, &p
);
1978 xor_hash
^= le64toh(o
->data
.hash
);
1979 items
[i
].object_offset
= htole64(p
);
1980 items
[i
].hash
= o
->data
.hash
;
1983 /* Order by the position on disk, in order to improve seek
1984 * times for rotating media. */
1985 qsort_safe(items
, n_iovec
, sizeof(EntryItem
), entry_item_cmp
);
1987 r
= journal_file_append_entry_internal(f
, ts
, xor_hash
, items
, n_iovec
, seqnum
, ret
, offset
);
1989 /* If the memory mapping triggered a SIGBUS then we return an
1990 * IO error and ignore the error code passed down to us, since
1991 * it is very likely just an effect of a nullified replacement
1994 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
1997 if (f
->post_change_timer
)
1998 schedule_post_change(f
);
2000 journal_file_post_change(f
);
2005 typedef struct ChainCacheItem
{
2006 uint64_t first
; /* the array at the beginning of the chain */
2007 uint64_t array
; /* the cached array */
2008 uint64_t begin
; /* the first item in the cached array */
2009 uint64_t total
; /* the total number of items in all arrays before this one in the chain */
2010 uint64_t last_index
; /* the last index we looked at, to optimize locality when bisecting */
2013 static void chain_cache_put(
2020 uint64_t last_index
) {
2023 /* If the chain item to cache for this chain is the
2024 * first one it's not worth caching anything */
2028 if (ordered_hashmap_size(h
) >= CHAIN_CACHE_MAX
) {
2029 ci
= ordered_hashmap_steal_first(h
);
2032 ci
= new(ChainCacheItem
, 1);
2039 if (ordered_hashmap_put(h
, &ci
->first
, ci
) < 0) {
2044 assert(ci
->first
== first
);
2049 ci
->last_index
= last_index
;
2052 static int generic_array_get(
2056 Object
**ret
, uint64_t *offset
) {
2059 uint64_t p
= 0, a
, t
= 0;
2067 /* Try the chain cache first */
2068 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
2069 if (ci
&& i
> ci
->total
) {
2078 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
2082 k
= journal_file_entry_array_n_items(o
);
2084 p
= le64toh(o
->entry_array
.items
[i
]);
2090 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
2096 /* Let's cache this item for the next invocation */
2097 chain_cache_put(f
->chain_cache
, ci
, first
, a
, le64toh(o
->entry_array
.items
[0]), t
, i
);
2099 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2112 static int generic_array_get_plus_one(
2117 Object
**ret
, uint64_t *offset
) {
2126 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, &o
);
2139 return generic_array_get(f
, first
, i
-1, ret
, offset
);
2148 static int generic_array_bisect(
2153 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
2154 direction_t direction
,
2159 uint64_t a
, p
, t
= 0, i
= 0, last_p
= 0, last_index
= (uint64_t) -1;
2160 bool subtract_one
= false;
2161 Object
*o
, *array
= NULL
;
2166 assert(test_object
);
2168 /* Start with the first array in the chain */
2171 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
2172 if (ci
&& n
> ci
->total
&& ci
->begin
!= 0) {
2173 /* Ah, we have iterated this bisection array chain
2174 * previously! Let's see if we can skip ahead in the
2175 * chain, as far as the last time. But we can't jump
2176 * backwards in the chain, so let's check that
2179 r
= test_object(f
, ci
->begin
, needle
);
2183 if (r
== TEST_LEFT
) {
2184 /* OK, what we are looking for is right of the
2185 * begin of this EntryArray, so let's jump
2186 * straight to previously cached array in the
2192 last_index
= ci
->last_index
;
2197 uint64_t left
, right
, k
, lp
;
2199 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &array
);
2203 k
= journal_file_entry_array_n_items(array
);
2209 lp
= p
= le64toh(array
->entry_array
.items
[i
]);
2213 r
= test_object(f
, p
, needle
);
2214 if (r
== -EBADMSG
) {
2215 log_debug_errno(r
, "Encountered invalid entry while bisecting, cutting algorithm short. (1)");
2222 if (r
== TEST_FOUND
)
2223 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2225 if (r
== TEST_RIGHT
) {
2229 if (last_index
!= (uint64_t) -1) {
2230 assert(last_index
<= right
);
2232 /* If we cached the last index we
2233 * looked at, let's try to not to jump
2234 * too wildly around and see if we can
2235 * limit the range to look at early to
2236 * the immediate neighbors of the last
2237 * index we looked at. */
2239 if (last_index
> 0) {
2240 uint64_t x
= last_index
- 1;
2242 p
= le64toh(array
->entry_array
.items
[x
]);
2246 r
= test_object(f
, p
, needle
);
2250 if (r
== TEST_FOUND
)
2251 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2253 if (r
== TEST_RIGHT
)
2259 if (last_index
< right
) {
2260 uint64_t y
= last_index
+ 1;
2262 p
= le64toh(array
->entry_array
.items
[y
]);
2266 r
= test_object(f
, p
, needle
);
2270 if (r
== TEST_FOUND
)
2271 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2273 if (r
== TEST_RIGHT
)
2281 if (left
== right
) {
2282 if (direction
== DIRECTION_UP
)
2283 subtract_one
= true;
2289 assert(left
< right
);
2290 i
= (left
+ right
) / 2;
2292 p
= le64toh(array
->entry_array
.items
[i
]);
2296 r
= test_object(f
, p
, needle
);
2297 if (r
== -EBADMSG
) {
2298 log_debug_errno(r
, "Encountered invalid entry while bisecting, cutting algorithm short. (2)");
2305 if (r
== TEST_FOUND
)
2306 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2308 if (r
== TEST_RIGHT
)
2316 if (direction
== DIRECTION_UP
) {
2318 subtract_one
= true;
2329 last_index
= (uint64_t) -1;
2330 a
= le64toh(array
->entry_array
.next_entry_array_offset
);
2336 if (subtract_one
&& t
== 0 && i
== 0)
2339 /* Let's cache this item for the next invocation */
2340 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
);
2342 if (subtract_one
&& i
== 0)
2344 else if (subtract_one
)
2345 p
= le64toh(array
->entry_array
.items
[i
-1]);
2347 p
= le64toh(array
->entry_array
.items
[i
]);
2349 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2360 *idx
= t
+ i
+ (subtract_one
? -1 : 0);
2365 static int generic_array_bisect_plus_one(
2371 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
2372 direction_t direction
,
2378 bool step_back
= false;
2382 assert(test_object
);
2387 /* This bisects the array in object 'first', but first checks
2389 r
= test_object(f
, extra
, needle
);
2393 if (r
== TEST_FOUND
)
2394 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2396 /* if we are looking with DIRECTION_UP then we need to first
2397 see if in the actual array there is a matching entry, and
2398 return the last one of that. But if there isn't any we need
2399 to return this one. Hence remember this, and return it
2402 step_back
= direction
== DIRECTION_UP
;
2404 if (r
== TEST_RIGHT
) {
2405 if (direction
== DIRECTION_DOWN
)
2411 r
= generic_array_bisect(f
, first
, n
-1, needle
, test_object
, direction
, ret
, offset
, idx
);
2413 if (r
== 0 && step_back
)
2422 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, &o
);
2438 _pure_
static int test_object_offset(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2444 else if (p
< needle
)
2450 static int test_object_seqnum(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2457 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2461 if (le64toh(o
->entry
.seqnum
) == needle
)
2463 else if (le64toh(o
->entry
.seqnum
) < needle
)
2469 int journal_file_move_to_entry_by_seqnum(
2472 direction_t direction
,
2478 return generic_array_bisect(f
,
2479 le64toh(f
->header
->entry_array_offset
),
2480 le64toh(f
->header
->n_entries
),
2487 static int test_object_realtime(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2494 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2498 if (le64toh(o
->entry
.realtime
) == needle
)
2500 else if (le64toh(o
->entry
.realtime
) < needle
)
2506 int journal_file_move_to_entry_by_realtime(
2509 direction_t direction
,
2515 return generic_array_bisect(f
,
2516 le64toh(f
->header
->entry_array_offset
),
2517 le64toh(f
->header
->n_entries
),
2519 test_object_realtime
,
2524 static int test_object_monotonic(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2531 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2535 if (le64toh(o
->entry
.monotonic
) == needle
)
2537 else if (le64toh(o
->entry
.monotonic
) < needle
)
2543 static int find_data_object_by_boot_id(
2549 char t
[STRLEN("_BOOT_ID=") + 32 + 1] = "_BOOT_ID=";
2551 sd_id128_to_string(boot_id
, t
+ 9);
2552 return journal_file_find_data_object(f
, t
, sizeof(t
) - 1, o
, b
);
2555 int journal_file_move_to_entry_by_monotonic(
2559 direction_t direction
,
2568 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, NULL
);
2574 return generic_array_bisect_plus_one(f
,
2575 le64toh(o
->data
.entry_offset
),
2576 le64toh(o
->data
.entry_array_offset
),
2577 le64toh(o
->data
.n_entries
),
2579 test_object_monotonic
,
2584 void journal_file_reset_location(JournalFile
*f
) {
2585 f
->location_type
= LOCATION_HEAD
;
2586 f
->current_offset
= 0;
2587 f
->current_seqnum
= 0;
2588 f
->current_realtime
= 0;
2589 f
->current_monotonic
= 0;
2590 zero(f
->current_boot_id
);
2591 f
->current_xor_hash
= 0;
2594 void journal_file_save_location(JournalFile
*f
, Object
*o
, uint64_t offset
) {
2595 f
->location_type
= LOCATION_SEEK
;
2596 f
->current_offset
= offset
;
2597 f
->current_seqnum
= le64toh(o
->entry
.seqnum
);
2598 f
->current_realtime
= le64toh(o
->entry
.realtime
);
2599 f
->current_monotonic
= le64toh(o
->entry
.monotonic
);
2600 f
->current_boot_id
= o
->entry
.boot_id
;
2601 f
->current_xor_hash
= le64toh(o
->entry
.xor_hash
);
2604 int journal_file_compare_locations(JournalFile
*af
, JournalFile
*bf
) {
2609 assert(af
->location_type
== LOCATION_SEEK
);
2610 assert(bf
->location_type
== LOCATION_SEEK
);
2612 /* If contents and timestamps match, these entries are
2613 * identical, even if the seqnum does not match */
2614 if (sd_id128_equal(af
->current_boot_id
, bf
->current_boot_id
) &&
2615 af
->current_monotonic
== bf
->current_monotonic
&&
2616 af
->current_realtime
== bf
->current_realtime
&&
2617 af
->current_xor_hash
== bf
->current_xor_hash
)
2620 if (sd_id128_equal(af
->header
->seqnum_id
, bf
->header
->seqnum_id
)) {
2622 /* If this is from the same seqnum source, compare
2624 if (af
->current_seqnum
< bf
->current_seqnum
)
2626 if (af
->current_seqnum
> bf
->current_seqnum
)
2629 /* Wow! This is weird, different data but the same
2630 * seqnums? Something is borked, but let's make the
2631 * best of it and compare by time. */
2634 if (sd_id128_equal(af
->current_boot_id
, bf
->current_boot_id
)) {
2636 /* If the boot id matches, compare monotonic time */
2637 if (af
->current_monotonic
< bf
->current_monotonic
)
2639 if (af
->current_monotonic
> bf
->current_monotonic
)
2643 /* Otherwise, compare UTC time */
2644 if (af
->current_realtime
< bf
->current_realtime
)
2646 if (af
->current_realtime
> bf
->current_realtime
)
2649 /* Finally, compare by contents */
2650 if (af
->current_xor_hash
< bf
->current_xor_hash
)
2652 if (af
->current_xor_hash
> bf
->current_xor_hash
)
2658 static int bump_array_index(uint64_t *i
, direction_t direction
, uint64_t n
) {
2660 /* Increase or decrease the specified index, in the right direction. */
2662 if (direction
== DIRECTION_DOWN
) {
2677 static bool check_properly_ordered(uint64_t new_offset
, uint64_t old_offset
, direction_t direction
) {
2679 /* Consider it an error if any of the two offsets is uninitialized */
2680 if (old_offset
== 0 || new_offset
== 0)
2683 /* If we go down, the new offset must be larger than the old one. */
2684 return direction
== DIRECTION_DOWN
?
2685 new_offset
> old_offset
:
2686 new_offset
< old_offset
;
2689 int journal_file_next_entry(
2692 direction_t direction
,
2693 Object
**ret
, uint64_t *offset
) {
2701 n
= le64toh(f
->header
->n_entries
);
2706 i
= direction
== DIRECTION_DOWN
? 0 : n
- 1;
2708 r
= generic_array_bisect(f
,
2709 le64toh(f
->header
->entry_array_offset
),
2710 le64toh(f
->header
->n_entries
),
2719 r
= bump_array_index(&i
, direction
, n
);
2724 /* And jump to it */
2726 r
= generic_array_get(f
,
2727 le64toh(f
->header
->entry_array_offset
),
2735 /* OK, so this entry is borked. Most likely some entry didn't get synced to disk properly, let's see if
2736 * the next one might work for us instead. */
2737 log_debug_errno(r
, "Entry item %" PRIu64
" is bad, skipping over it.", i
);
2739 r
= bump_array_index(&i
, direction
, n
);
2744 /* Ensure our array is properly ordered. */
2745 if (p
> 0 && !check_properly_ordered(ofs
, p
, direction
)) {
2746 log_debug("%s: entry array not properly ordered at entry %" PRIu64
, f
->path
, i
);
2756 int journal_file_next_entry_for_data(
2758 Object
*o
, uint64_t p
,
2759 uint64_t data_offset
,
2760 direction_t direction
,
2761 Object
**ret
, uint64_t *offset
) {
2768 assert(p
> 0 || !o
);
2770 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2774 n
= le64toh(d
->data
.n_entries
);
2779 i
= direction
== DIRECTION_DOWN
? 0 : n
- 1;
2781 if (o
->object
.type
!= OBJECT_ENTRY
)
2784 r
= generic_array_bisect_plus_one(f
,
2785 le64toh(d
->data
.entry_offset
),
2786 le64toh(d
->data
.entry_array_offset
),
2787 le64toh(d
->data
.n_entries
),
2797 r
= bump_array_index(&i
, direction
, n
);
2803 r
= generic_array_get_plus_one(f
,
2804 le64toh(d
->data
.entry_offset
),
2805 le64toh(d
->data
.entry_array_offset
),
2813 log_debug_errno(r
, "Data entry item %" PRIu64
" is bad, skipping over it.", i
);
2815 r
= bump_array_index(&i
, direction
, n
);
2820 /* Ensure our array is properly ordered. */
2821 if (p
> 0 && check_properly_ordered(ofs
, p
, direction
)) {
2822 log_debug("%s data entry array not properly ordered at entry %" PRIu64
, f
->path
, i
);
2832 int journal_file_move_to_entry_by_offset_for_data(
2834 uint64_t data_offset
,
2836 direction_t direction
,
2837 Object
**ret
, uint64_t *offset
) {
2844 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2848 return generic_array_bisect_plus_one(f
,
2849 le64toh(d
->data
.entry_offset
),
2850 le64toh(d
->data
.entry_array_offset
),
2851 le64toh(d
->data
.n_entries
),
2858 int journal_file_move_to_entry_by_monotonic_for_data(
2860 uint64_t data_offset
,
2863 direction_t direction
,
2864 Object
**ret
, uint64_t *offset
) {
2872 /* First, seek by time */
2873 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, &b
);
2879 r
= generic_array_bisect_plus_one(f
,
2880 le64toh(o
->data
.entry_offset
),
2881 le64toh(o
->data
.entry_array_offset
),
2882 le64toh(o
->data
.n_entries
),
2884 test_object_monotonic
,
2890 /* And now, continue seeking until we find an entry that
2891 * exists in both bisection arrays */
2897 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2901 r
= generic_array_bisect_plus_one(f
,
2902 le64toh(d
->data
.entry_offset
),
2903 le64toh(d
->data
.entry_array_offset
),
2904 le64toh(d
->data
.n_entries
),
2912 r
= journal_file_move_to_object(f
, OBJECT_DATA
, b
, &o
);
2916 r
= generic_array_bisect_plus_one(f
,
2917 le64toh(o
->data
.entry_offset
),
2918 le64toh(o
->data
.entry_array_offset
),
2919 le64toh(o
->data
.n_entries
),
2941 int journal_file_move_to_entry_by_seqnum_for_data(
2943 uint64_t data_offset
,
2945 direction_t direction
,
2946 Object
**ret
, uint64_t *offset
) {
2953 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2957 return generic_array_bisect_plus_one(f
,
2958 le64toh(d
->data
.entry_offset
),
2959 le64toh(d
->data
.entry_array_offset
),
2960 le64toh(d
->data
.n_entries
),
2967 int journal_file_move_to_entry_by_realtime_for_data(
2969 uint64_t data_offset
,
2971 direction_t direction
,
2972 Object
**ret
, uint64_t *offset
) {
2979 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2983 return generic_array_bisect_plus_one(f
,
2984 le64toh(d
->data
.entry_offset
),
2985 le64toh(d
->data
.entry_array_offset
),
2986 le64toh(d
->data
.n_entries
),
2988 test_object_realtime
,
2993 void journal_file_dump(JournalFile
*f
) {
3001 journal_file_print_header(f
);
3003 p
= le64toh(f
->header
->header_size
);
3005 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &o
);
3009 switch (o
->object
.type
) {
3012 printf("Type: OBJECT_UNUSED\n");
3016 printf("Type: OBJECT_DATA\n");
3020 printf("Type: OBJECT_FIELD\n");
3024 printf("Type: OBJECT_ENTRY seqnum=%"PRIu64
" monotonic=%"PRIu64
" realtime=%"PRIu64
"\n",
3025 le64toh(o
->entry
.seqnum
),
3026 le64toh(o
->entry
.monotonic
),
3027 le64toh(o
->entry
.realtime
));
3030 case OBJECT_FIELD_HASH_TABLE
:
3031 printf("Type: OBJECT_FIELD_HASH_TABLE\n");
3034 case OBJECT_DATA_HASH_TABLE
:
3035 printf("Type: OBJECT_DATA_HASH_TABLE\n");
3038 case OBJECT_ENTRY_ARRAY
:
3039 printf("Type: OBJECT_ENTRY_ARRAY\n");
3043 printf("Type: OBJECT_TAG seqnum=%"PRIu64
" epoch=%"PRIu64
"\n",
3044 le64toh(o
->tag
.seqnum
),
3045 le64toh(o
->tag
.epoch
));
3049 printf("Type: unknown (%i)\n", o
->object
.type
);
3053 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
)
3054 printf("Flags: %s\n",
3055 object_compressed_to_string(o
->object
.flags
& OBJECT_COMPRESSION_MASK
));
3057 if (p
== le64toh(f
->header
->tail_object_offset
))
3060 p
= p
+ ALIGN64(le64toh(o
->object
.size
));
3065 log_error("File corrupt");
3068 static const char* format_timestamp_safe(char *buf
, size_t l
, usec_t t
) {
3071 x
= format_timestamp(buf
, l
, t
);
3077 void journal_file_print_header(JournalFile
*f
) {
3078 char a
[33], b
[33], c
[33], d
[33];
3079 char x
[FORMAT_TIMESTAMP_MAX
], y
[FORMAT_TIMESTAMP_MAX
], z
[FORMAT_TIMESTAMP_MAX
];
3081 char bytes
[FORMAT_BYTES_MAX
];
3086 printf("File Path: %s\n"
3090 "Sequential Number ID: %s\n"
3092 "Compatible Flags:%s%s\n"
3093 "Incompatible Flags:%s%s%s\n"
3094 "Header size: %"PRIu64
"\n"
3095 "Arena size: %"PRIu64
"\n"
3096 "Data Hash Table Size: %"PRIu64
"\n"
3097 "Field Hash Table Size: %"PRIu64
"\n"
3098 "Rotate Suggested: %s\n"
3099 "Head Sequential Number: %"PRIu64
" (%"PRIx64
")\n"
3100 "Tail Sequential Number: %"PRIu64
" (%"PRIx64
")\n"
3101 "Head Realtime Timestamp: %s (%"PRIx64
")\n"
3102 "Tail Realtime Timestamp: %s (%"PRIx64
")\n"
3103 "Tail Monotonic Timestamp: %s (%"PRIx64
")\n"
3104 "Objects: %"PRIu64
"\n"
3105 "Entry Objects: %"PRIu64
"\n",
3107 sd_id128_to_string(f
->header
->file_id
, a
),
3108 sd_id128_to_string(f
->header
->machine_id
, b
),
3109 sd_id128_to_string(f
->header
->boot_id
, c
),
3110 sd_id128_to_string(f
->header
->seqnum_id
, d
),
3111 f
->header
->state
== STATE_OFFLINE
? "OFFLINE" :
3112 f
->header
->state
== STATE_ONLINE
? "ONLINE" :
3113 f
->header
->state
== STATE_ARCHIVED
? "ARCHIVED" : "UNKNOWN",
3114 JOURNAL_HEADER_SEALED(f
->header
) ? " SEALED" : "",
3115 (le32toh(f
->header
->compatible_flags
) & ~HEADER_COMPATIBLE_ANY
) ? " ???" : "",
3116 JOURNAL_HEADER_COMPRESSED_XZ(f
->header
) ? " COMPRESSED-XZ" : "",
3117 JOURNAL_HEADER_COMPRESSED_LZ4(f
->header
) ? " COMPRESSED-LZ4" : "",
3118 (le32toh(f
->header
->incompatible_flags
) & ~HEADER_INCOMPATIBLE_ANY
) ? " ???" : "",
3119 le64toh(f
->header
->header_size
),
3120 le64toh(f
->header
->arena_size
),
3121 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
3122 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
),
3123 yes_no(journal_file_rotate_suggested(f
, 0)),
3124 le64toh(f
->header
->head_entry_seqnum
), le64toh(f
->header
->head_entry_seqnum
),
3125 le64toh(f
->header
->tail_entry_seqnum
), le64toh(f
->header
->tail_entry_seqnum
),
3126 format_timestamp_safe(x
, sizeof(x
), le64toh(f
->header
->head_entry_realtime
)), le64toh(f
->header
->head_entry_realtime
),
3127 format_timestamp_safe(y
, sizeof(y
), le64toh(f
->header
->tail_entry_realtime
)), le64toh(f
->header
->tail_entry_realtime
),
3128 format_timespan(z
, sizeof(z
), le64toh(f
->header
->tail_entry_monotonic
), USEC_PER_MSEC
), le64toh(f
->header
->tail_entry_monotonic
),
3129 le64toh(f
->header
->n_objects
),
3130 le64toh(f
->header
->n_entries
));
3132 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
3133 printf("Data Objects: %"PRIu64
"\n"
3134 "Data Hash Table Fill: %.1f%%\n",
3135 le64toh(f
->header
->n_data
),
3136 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))));
3138 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
3139 printf("Field Objects: %"PRIu64
"\n"
3140 "Field Hash Table Fill: %.1f%%\n",
3141 le64toh(f
->header
->n_fields
),
3142 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))));
3144 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_tags
))
3145 printf("Tag Objects: %"PRIu64
"\n",
3146 le64toh(f
->header
->n_tags
));
3147 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
3148 printf("Entry Array Objects: %"PRIu64
"\n",
3149 le64toh(f
->header
->n_entry_arrays
));
3151 if (fstat(f
->fd
, &st
) >= 0)
3152 printf("Disk usage: %s\n", format_bytes(bytes
, sizeof(bytes
), (uint64_t) st
.st_blocks
* 512ULL));
3155 static int journal_file_warn_btrfs(JournalFile
*f
) {
3161 /* Before we write anything, check if the COW logic is turned
3162 * off on btrfs. Given our write pattern that is quite
3163 * unfriendly to COW file systems this should greatly improve
3164 * performance on COW file systems, such as btrfs, at the
3165 * expense of data integrity features (which shouldn't be too
3166 * bad, given that we do our own checksumming). */
3168 r
= btrfs_is_filesystem(f
->fd
);
3170 return log_warning_errno(r
, "Failed to determine if journal is on btrfs: %m");
3174 r
= read_attr_fd(f
->fd
, &attrs
);
3176 return log_warning_errno(r
, "Failed to read file attributes: %m");
3178 if (attrs
& FS_NOCOW_FL
) {
3179 log_debug("Detected btrfs file system with copy-on-write disabled, all is good.");
3183 log_notice("Creating journal file %s on a btrfs file system, and copy-on-write is enabled. "
3184 "This is likely to slow down journal access substantially, please consider turning "
3185 "off the copy-on-write file attribute on the journal directory, using chattr +C.", f
->path
);
3190 int journal_file_open(
3196 uint64_t compress_threshold_bytes
,
3198 JournalMetrics
*metrics
,
3199 MMapCache
*mmap_cache
,
3200 Set
*deferred_closes
,
3201 JournalFile
*template,
3202 JournalFile
**ret
) {
3204 bool newly_created
= false;
3208 char bytes
[FORMAT_BYTES_MAX
];
3211 assert(fd
>= 0 || fname
);
3213 if (!IN_SET((flags
& O_ACCMODE
), O_RDONLY
, O_RDWR
))
3216 if (fname
&& (flags
& O_CREAT
) && !endswith(fname
, ".journal"))
3219 f
= new0(JournalFile
, 1);
3227 f
->prot
= prot_from_flags(flags
);
3228 f
->writable
= (flags
& O_ACCMODE
) != O_RDONLY
;
3230 f
->compress_lz4
= compress
;
3232 f
->compress_xz
= compress
;
3235 if (compress_threshold_bytes
== (uint64_t) -1)
3236 f
->compress_threshold_bytes
= DEFAULT_COMPRESS_THRESHOLD
;
3238 f
->compress_threshold_bytes
= MAX(MIN_COMPRESS_THRESHOLD
, compress_threshold_bytes
);
3244 log_debug("Journal effective settings seal=%s compress=%s compress_threshold_bytes=%s",
3245 yes_no(f
->seal
), yes_no(JOURNAL_FILE_COMPRESS(f
)),
3246 format_bytes(bytes
, sizeof(bytes
), f
->compress_threshold_bytes
));
3249 f
->mmap
= mmap_cache_ref(mmap_cache
);
3251 f
->mmap
= mmap_cache_new();
3259 f
->path
= strdup(fname
);
3267 /* If we don't know the path, fill in something explanatory and vaguely useful */
3268 if (asprintf(&f
->path
, "/proc/self/%i", fd
) < 0) {
3274 f
->chain_cache
= ordered_hashmap_new(&uint64_hash_ops
);
3275 if (!f
->chain_cache
) {
3281 /* We pass O_NONBLOCK here, so that in case somebody pointed us to some character device node or FIFO
3282 * or so, we likely fail quickly than block for long. For regular files O_NONBLOCK has no effect, hence
3283 * it doesn't hurt in that case. */
3285 f
->fd
= open(f
->path
, f
->flags
|O_CLOEXEC
|O_NONBLOCK
, f
->mode
);
3291 /* fds we opened here by us should also be closed by us. */
3294 r
= fd_nonblock(f
->fd
, false);
3299 f
->cache_fd
= mmap_cache_add_fd(f
->mmap
, f
->fd
);
3305 r
= journal_file_fstat(f
);
3309 if (f
->last_stat
.st_size
== 0 && f
->writable
) {
3311 (void) journal_file_warn_btrfs(f
);
3313 /* Let's attach the creation time to the journal file, so that the vacuuming code knows the age of this
3314 * file even if the file might end up corrupted one day... Ideally we'd just use the creation time many
3315 * file systems maintain for each file, but the API to query this is very new, hence let's emulate this
3316 * via extended attributes. If extended attributes are not supported we'll just skip this, and rely
3317 * solely on mtime/atime/ctime of the file. */
3318 (void) fd_setcrtime(f
->fd
, 0);
3321 /* Try to load the FSPRG state, and if we can't, then
3322 * just don't do sealing */
3324 r
= journal_file_fss_load(f
);
3330 r
= journal_file_init_header(f
, template);
3334 r
= journal_file_fstat(f
);
3338 newly_created
= true;
3341 if (f
->last_stat
.st_size
< (off_t
) HEADER_SIZE_MIN
) {
3346 r
= mmap_cache_get(f
->mmap
, f
->cache_fd
, f
->prot
, CONTEXT_HEADER
, true, 0, PAGE_ALIGN(sizeof(Header
)), &f
->last_stat
, &h
, NULL
);
3352 if (!newly_created
) {
3353 set_clear_with_destructor(deferred_closes
, journal_file_close
);
3355 r
= journal_file_verify_header(f
);
3361 if (!newly_created
&& f
->writable
) {
3362 r
= journal_file_fss_load(f
);
3370 journal_default_metrics(metrics
, f
->fd
);
3371 f
->metrics
= *metrics
;
3372 } else if (template)
3373 f
->metrics
= template->metrics
;
3375 r
= journal_file_refresh_header(f
);
3381 r
= journal_file_hmac_setup(f
);
3386 if (newly_created
) {
3387 r
= journal_file_setup_field_hash_table(f
);
3391 r
= journal_file_setup_data_hash_table(f
);
3396 r
= journal_file_append_first_tag(f
);
3402 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
)) {
3407 if (template && template->post_change_timer
) {
3408 r
= journal_file_enable_post_change_timer(
3410 sd_event_source_get_event(template->post_change_timer
),
3411 template->post_change_timer_period
);
3417 /* The file is opened now successfully, thus we take possession of any passed in fd. */
3424 if (f
->cache_fd
&& mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
3427 (void) journal_file_close(f
);
3432 int journal_file_rotate(JournalFile
**f
, bool compress
, uint64_t compress_threshold_bytes
, bool seal
, Set
*deferred_closes
) {
3433 _cleanup_free_
char *p
= NULL
;
3435 JournalFile
*old_file
, *new_file
= NULL
;
3443 if (!old_file
->writable
)
3446 /* Is this a journal file that was passed to us as fd? If so, we synthesized a path name for it, and we refuse
3447 * rotation, since we don't know the actual path, and couldn't rename the file hence. */
3448 if (path_startswith(old_file
->path
, "/proc/self/fd"))
3451 if (!endswith(old_file
->path
, ".journal"))
3454 l
= strlen(old_file
->path
);
3455 r
= asprintf(&p
, "%.*s@" SD_ID128_FORMAT_STR
"-%016"PRIx64
"-%016"PRIx64
".journal",
3456 (int) l
- 8, old_file
->path
,
3457 SD_ID128_FORMAT_VAL(old_file
->header
->seqnum_id
),
3458 le64toh((*f
)->header
->head_entry_seqnum
),
3459 le64toh((*f
)->header
->head_entry_realtime
));
3463 /* Try to rename the file to the archived version. If the file
3464 * already was deleted, we'll get ENOENT, let's ignore that
3466 r
= rename(old_file
->path
, p
);
3467 if (r
< 0 && errno
!= ENOENT
)
3470 /* Sync the rename to disk */
3471 (void) fsync_directory_of_file(old_file
->fd
);
3473 /* Set as archive so offlining commits w/state=STATE_ARCHIVED.
3474 * Previously we would set old_file->header->state to STATE_ARCHIVED directly here,
3475 * but journal_file_set_offline() short-circuits when state != STATE_ONLINE, which
3476 * would result in the rotated journal never getting fsync() called before closing.
3477 * Now we simply queue the archive state by setting an archive bit, leaving the state
3478 * as STATE_ONLINE so proper offlining occurs. */
3479 old_file
->archive
= true;
3481 /* Currently, btrfs is not very good with out write patterns
3482 * and fragments heavily. Let's defrag our journal files when
3483 * we archive them */
3484 old_file
->defrag_on_close
= true;
3486 r
= journal_file_open(-1, old_file
->path
, old_file
->flags
, old_file
->mode
, compress
,
3487 compress_threshold_bytes
, seal
, NULL
, old_file
->mmap
, deferred_closes
,
3488 old_file
, &new_file
);
3490 if (deferred_closes
&&
3491 set_put(deferred_closes
, old_file
) >= 0)
3492 (void) journal_file_set_offline(old_file
, false);
3494 (void) journal_file_close(old_file
);
3500 int journal_file_open_reliably(
3505 uint64_t compress_threshold_bytes
,
3507 JournalMetrics
*metrics
,
3508 MMapCache
*mmap_cache
,
3509 Set
*deferred_closes
,
3510 JournalFile
*template,
3511 JournalFile
**ret
) {
3515 _cleanup_free_
char *p
= NULL
;
3517 r
= journal_file_open(-1, fname
, flags
, mode
, compress
, compress_threshold_bytes
, seal
, metrics
, mmap_cache
,
3518 deferred_closes
, template, ret
);
3520 -EBADMSG
, /* Corrupted */
3521 -ENODATA
, /* Truncated */
3522 -EHOSTDOWN
, /* Other machine */
3523 -EPROTONOSUPPORT
, /* Incompatible feature */
3524 -EBUSY
, /* Unclean shutdown */
3525 -ESHUTDOWN
, /* Already archived */
3526 -EIO
, /* IO error, including SIGBUS on mmap */
3527 -EIDRM
, /* File has been deleted */
3528 -ETXTBSY
)) /* File is from the future */
3531 if ((flags
& O_ACCMODE
) == O_RDONLY
)
3534 if (!(flags
& O_CREAT
))
3537 if (!endswith(fname
, ".journal"))
3540 /* The file is corrupted. Rotate it away and try it again (but only once) */
3543 if (asprintf(&p
, "%.*s@%016"PRIx64
"-%016"PRIx64
".journal~",
3545 now(CLOCK_REALTIME
),
3549 if (rename(fname
, p
) < 0)
3552 /* btrfs doesn't cope well with our write pattern and
3553 * fragments heavily. Let's defrag all files we rotate */
3555 (void) chattr_path(p
, 0, FS_NOCOW_FL
);
3556 (void) btrfs_defrag(p
);
3558 log_warning_errno(r
, "File %s corrupted or uncleanly shut down, renaming and replacing.", fname
);
3560 return journal_file_open(-1, fname
, flags
, mode
, compress
, compress_threshold_bytes
, seal
, metrics
, mmap_cache
,
3561 deferred_closes
, template, ret
);
3564 int journal_file_copy_entry(JournalFile
*from
, JournalFile
*to
, Object
*o
, uint64_t p
, uint64_t *seqnum
, Object
**ret
, uint64_t *offset
) {
3566 uint64_t q
, xor_hash
= 0;
3579 ts
.monotonic
= le64toh(o
->entry
.monotonic
);
3580 ts
.realtime
= le64toh(o
->entry
.realtime
);
3582 n
= journal_file_entry_n_items(o
);
3583 /* alloca() can't take 0, hence let's allocate at least one */
3584 items
= newa(EntryItem
, MAX(1u, n
));
3586 for (i
= 0; i
< n
; i
++) {
3593 q
= le64toh(o
->entry
.items
[i
].object_offset
);
3594 le_hash
= o
->entry
.items
[i
].hash
;
3596 r
= journal_file_move_to_object(from
, OBJECT_DATA
, q
, &o
);
3600 if (le_hash
!= o
->data
.hash
)
3603 l
= le64toh(o
->object
.size
) - offsetof(Object
, data
.payload
);
3606 /* We hit the limit on 32bit machines */
3607 if ((uint64_t) t
!= l
)
3610 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
) {
3611 #if HAVE_XZ || HAVE_LZ4
3614 r
= decompress_blob(o
->object
.flags
& OBJECT_COMPRESSION_MASK
,
3615 o
->data
.payload
, l
, &from
->compress_buffer
, &from
->compress_buffer_size
, &rsize
, 0);
3619 data
= from
->compress_buffer
;
3622 return -EPROTONOSUPPORT
;
3625 data
= o
->data
.payload
;
3627 r
= journal_file_append_data(to
, data
, l
, &u
, &h
);
3631 xor_hash
^= le64toh(u
->data
.hash
);
3632 items
[i
].object_offset
= htole64(h
);
3633 items
[i
].hash
= u
->data
.hash
;
3635 r
= journal_file_move_to_object(from
, OBJECT_ENTRY
, p
, &o
);
3640 r
= journal_file_append_entry_internal(to
, &ts
, xor_hash
, items
, n
, seqnum
, ret
, offset
);
3642 if (mmap_cache_got_sigbus(to
->mmap
, to
->cache_fd
))
3648 void journal_reset_metrics(JournalMetrics
*m
) {
3651 /* Set everything to "pick automatic values". */
3653 *m
= (JournalMetrics
) {
3654 .min_use
= (uint64_t) -1,
3655 .max_use
= (uint64_t) -1,
3656 .min_size
= (uint64_t) -1,
3657 .max_size
= (uint64_t) -1,
3658 .keep_free
= (uint64_t) -1,
3659 .n_max_files
= (uint64_t) -1,
3663 void journal_default_metrics(JournalMetrics
*m
, int fd
) {
3664 char a
[FORMAT_BYTES_MAX
], b
[FORMAT_BYTES_MAX
], c
[FORMAT_BYTES_MAX
], d
[FORMAT_BYTES_MAX
], e
[FORMAT_BYTES_MAX
];
3671 if (fstatvfs(fd
, &ss
) >= 0)
3672 fs_size
= ss
.f_frsize
* ss
.f_blocks
;
3674 log_debug_errno(errno
, "Failed to determine disk size: %m");
3678 if (m
->max_use
== (uint64_t) -1) {
3681 m
->max_use
= PAGE_ALIGN(fs_size
/ 10); /* 10% of file system size */
3683 if (m
->max_use
> DEFAULT_MAX_USE_UPPER
)
3684 m
->max_use
= DEFAULT_MAX_USE_UPPER
;
3686 if (m
->max_use
< DEFAULT_MAX_USE_LOWER
)
3687 m
->max_use
= DEFAULT_MAX_USE_LOWER
;
3689 m
->max_use
= DEFAULT_MAX_USE_LOWER
;
3691 m
->max_use
= PAGE_ALIGN(m
->max_use
);
3693 if (m
->max_use
!= 0 && m
->max_use
< JOURNAL_FILE_SIZE_MIN
*2)
3694 m
->max_use
= JOURNAL_FILE_SIZE_MIN
*2;
3697 if (m
->min_use
== (uint64_t) -1)
3698 m
->min_use
= DEFAULT_MIN_USE
;
3700 if (m
->min_use
> m
->max_use
)
3701 m
->min_use
= m
->max_use
;
3703 if (m
->max_size
== (uint64_t) -1) {
3704 m
->max_size
= PAGE_ALIGN(m
->max_use
/ 8); /* 8 chunks */
3706 if (m
->max_size
> DEFAULT_MAX_SIZE_UPPER
)
3707 m
->max_size
= DEFAULT_MAX_SIZE_UPPER
;
3709 m
->max_size
= PAGE_ALIGN(m
->max_size
);
3711 if (m
->max_size
!= 0) {
3712 if (m
->max_size
< JOURNAL_FILE_SIZE_MIN
)
3713 m
->max_size
= JOURNAL_FILE_SIZE_MIN
;
3715 if (m
->max_use
!= 0 && m
->max_size
*2 > m
->max_use
)
3716 m
->max_use
= m
->max_size
*2;
3719 if (m
->min_size
== (uint64_t) -1)
3720 m
->min_size
= JOURNAL_FILE_SIZE_MIN
;
3722 m
->min_size
= PAGE_ALIGN(m
->min_size
);
3724 if (m
->min_size
< JOURNAL_FILE_SIZE_MIN
)
3725 m
->min_size
= JOURNAL_FILE_SIZE_MIN
;
3727 if (m
->max_size
!= 0 && m
->min_size
> m
->max_size
)
3728 m
->max_size
= m
->min_size
;
3731 if (m
->keep_free
== (uint64_t) -1) {
3734 m
->keep_free
= PAGE_ALIGN(fs_size
* 3 / 20); /* 15% of file system size */
3736 if (m
->keep_free
> DEFAULT_KEEP_FREE_UPPER
)
3737 m
->keep_free
= DEFAULT_KEEP_FREE_UPPER
;
3740 m
->keep_free
= DEFAULT_KEEP_FREE
;
3743 if (m
->n_max_files
== (uint64_t) -1)
3744 m
->n_max_files
= DEFAULT_N_MAX_FILES
;
3746 log_debug("Fixed min_use=%s max_use=%s max_size=%s min_size=%s keep_free=%s n_max_files=%" PRIu64
,
3747 format_bytes(a
, sizeof(a
), m
->min_use
),
3748 format_bytes(b
, sizeof(b
), m
->max_use
),
3749 format_bytes(c
, sizeof(c
), m
->max_size
),
3750 format_bytes(d
, sizeof(d
), m
->min_size
),
3751 format_bytes(e
, sizeof(e
), m
->keep_free
),
3755 int journal_file_get_cutoff_realtime_usec(JournalFile
*f
, usec_t
*from
, usec_t
*to
) {
3761 if (f
->header
->head_entry_realtime
== 0)
3764 *from
= le64toh(f
->header
->head_entry_realtime
);
3768 if (f
->header
->tail_entry_realtime
== 0)
3771 *to
= le64toh(f
->header
->tail_entry_realtime
);
3777 int journal_file_get_cutoff_monotonic_usec(JournalFile
*f
, sd_id128_t boot_id
, usec_t
*from
, usec_t
*to
) {
3785 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, &p
);
3789 if (le64toh(o
->data
.n_entries
) <= 0)
3793 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, le64toh(o
->data
.entry_offset
), &o
);
3797 *from
= le64toh(o
->entry
.monotonic
);
3801 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
3805 r
= generic_array_get_plus_one(f
,
3806 le64toh(o
->data
.entry_offset
),
3807 le64toh(o
->data
.entry_array_offset
),
3808 le64toh(o
->data
.n_entries
)-1,
3813 *to
= le64toh(o
->entry
.monotonic
);
3819 bool journal_file_rotate_suggested(JournalFile
*f
, usec_t max_file_usec
) {
3823 /* If we gained new header fields we gained new features,
3824 * hence suggest a rotation */
3825 if (le64toh(f
->header
->header_size
) < sizeof(Header
)) {
3826 log_debug("%s uses an outdated header, suggesting rotation.", f
->path
);
3830 /* Let's check if the hash tables grew over a certain fill
3831 * level (75%, borrowing this value from Java's hash table
3832 * implementation), and if so suggest a rotation. To calculate
3833 * the fill level we need the n_data field, which only exists
3834 * in newer versions. */
3836 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
3837 if (le64toh(f
->header
->n_data
) * 4ULL > (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
3838 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.",
3840 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))),
3841 le64toh(f
->header
->n_data
),
3842 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
3843 (unsigned long long) f
->last_stat
.st_size
,
3844 f
->last_stat
.st_size
/ le64toh(f
->header
->n_data
));
3848 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
3849 if (le64toh(f
->header
->n_fields
) * 4ULL > (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
3850 log_debug("Field hash table of %s has a fill level at %.1f (%"PRIu64
" of %"PRIu64
" items), suggesting rotation.",
3852 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))),
3853 le64toh(f
->header
->n_fields
),
3854 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
));
3858 /* Are the data objects properly indexed by field objects? */
3859 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
) &&
3860 JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
) &&
3861 le64toh(f
->header
->n_data
) > 0 &&
3862 le64toh(f
->header
->n_fields
) == 0)
3865 if (max_file_usec
> 0) {
3868 h
= le64toh(f
->header
->head_entry_realtime
);
3869 t
= now(CLOCK_REALTIME
);
3871 if (h
> 0 && t
> h
+ max_file_usec
)