2 This file is part of systemd.
4 Copyright 2011 Lennart Poettering
6 systemd is free software; you can redistribute it and/or modify it
7 under the terms of the GNU Lesser General Public License as published by
8 the Free Software Foundation; either version 2.1 of the License, or
9 (at your option) any later version.
11 systemd is distributed in the hope that it will be useful, but
12 WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 Lesser General Public License for more details.
16 You should have received a copy of the GNU Lesser General Public License
17 along with systemd; If not, see <http://www.gnu.org/licenses/>.
26 #include <sys/statvfs.h>
30 #include "alloc-util.h"
31 #include "btrfs-util.h"
32 #include "chattr-util.h"
35 #include "journal-authenticate.h"
36 #include "journal-def.h"
37 #include "journal-file.h"
39 #include "parse-util.h"
40 #include "path-util.h"
41 #include "random-util.h"
44 #include "string-util.h"
46 #include "xattr-util.h"
48 #define DEFAULT_DATA_HASH_TABLE_SIZE (2047ULL*sizeof(HashItem))
49 #define DEFAULT_FIELD_HASH_TABLE_SIZE (333ULL*sizeof(HashItem))
51 #define COMPRESSION_SIZE_THRESHOLD (512ULL)
53 /* This is the minimum journal file size */
54 #define JOURNAL_FILE_SIZE_MIN (512ULL*1024ULL) /* 512 KiB */
56 /* These are the lower and upper bounds if we deduce the max_use value
57 * from the file system size */
58 #define DEFAULT_MAX_USE_LOWER (1ULL*1024ULL*1024ULL) /* 1 MiB */
59 #define DEFAULT_MAX_USE_UPPER (4ULL*1024ULL*1024ULL*1024ULL) /* 4 GiB */
61 /* This is the default minimal use limit, how much we'll use even if keep_free suggests otherwise. */
62 #define DEFAULT_MIN_USE (1ULL*1024ULL*1024ULL) /* 1 MiB */
64 /* This is the upper bound if we deduce max_size from max_use */
65 #define DEFAULT_MAX_SIZE_UPPER (128ULL*1024ULL*1024ULL) /* 128 MiB */
67 /* This is the upper bound if we deduce the keep_free value from the
69 #define DEFAULT_KEEP_FREE_UPPER (4ULL*1024ULL*1024ULL*1024ULL) /* 4 GiB */
71 /* This is the keep_free value when we can't determine the system
73 #define DEFAULT_KEEP_FREE (1024ULL*1024ULL) /* 1 MB */
75 /* This is the default maximum number of journal files to keep around. */
76 #define DEFAULT_N_MAX_FILES (100)
78 /* n_data was the first entry we added after the initial file format design */
79 #define HEADER_SIZE_MIN ALIGN64(offsetof(Header, n_data))
81 /* How many entries to keep in the entry array chain cache at max */
82 #define CHAIN_CACHE_MAX 20
84 /* How much to increase the journal file size at once each time we allocate something new. */
85 #define FILE_SIZE_INCREASE (8ULL*1024ULL*1024ULL) /* 8MB */
87 /* Reread fstat() of the file for detecting deletions at least this often */
88 #define LAST_STAT_REFRESH_USEC (5*USEC_PER_SEC)
90 /* The mmap context to use for the header we pick as one above the last defined typed */
91 #define CONTEXT_HEADER _OBJECT_TYPE_MAX
93 /* This may be called from a separate thread to prevent blocking the caller for the duration of fsync().
94 * As a result we use atomic operations on f->offline_state for inter-thread communications with
95 * journal_file_set_offline() and journal_file_set_online(). */
96 static void journal_file_set_offline_internal(JournalFile
*f
) {
102 switch (f
->offline_state
) {
104 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_CANCEL
, OFFLINE_DONE
))
108 case OFFLINE_AGAIN_FROM_SYNCING
:
109 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_AGAIN_FROM_SYNCING
, OFFLINE_SYNCING
))
113 case OFFLINE_AGAIN_FROM_OFFLINING
:
114 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_AGAIN_FROM_OFFLINING
, OFFLINE_SYNCING
))
118 case OFFLINE_SYNCING
:
121 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_SYNCING
, OFFLINE_OFFLINING
))
124 f
->header
->state
= f
->archive
? STATE_ARCHIVED
: STATE_OFFLINE
;
128 case OFFLINE_OFFLINING
:
129 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_OFFLINING
, OFFLINE_DONE
))
137 log_debug("OFFLINE_JOINED unexpected offline state for journal_file_set_offline_internal()");
143 static void * journal_file_set_offline_thread(void *arg
) {
144 JournalFile
*f
= arg
;
146 journal_file_set_offline_internal(f
);
151 static int journal_file_set_offline_thread_join(JournalFile
*f
) {
156 if (f
->offline_state
== OFFLINE_JOINED
)
159 r
= pthread_join(f
->offline_thread
, NULL
);
163 f
->offline_state
= OFFLINE_JOINED
;
165 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
171 /* Trigger a restart if the offline thread is mid-flight in a restartable state. */
172 static bool journal_file_set_offline_try_restart(JournalFile
*f
) {
174 switch (f
->offline_state
) {
175 case OFFLINE_AGAIN_FROM_SYNCING
:
176 case OFFLINE_AGAIN_FROM_OFFLINING
:
180 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_CANCEL
, OFFLINE_AGAIN_FROM_SYNCING
))
184 case OFFLINE_SYNCING
:
185 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_SYNCING
, OFFLINE_AGAIN_FROM_SYNCING
))
189 case OFFLINE_OFFLINING
:
190 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_OFFLINING
, OFFLINE_AGAIN_FROM_OFFLINING
))
200 /* Sets a journal offline.
202 * If wait is false then an offline is dispatched in a separate thread for a
203 * subsequent journal_file_set_offline() or journal_file_set_online() of the
204 * same journal to synchronize with.
206 * If wait is true, then either an existing offline thread will be restarted
207 * and joined, or if none exists the offline is simply performed in this
208 * context without involving another thread.
210 int journal_file_set_offline(JournalFile
*f
, bool wait
) {
219 if (!(f
->fd
>= 0 && f
->header
))
222 /* An offlining journal is implicitly online and may modify f->header->state,
223 * we must also join any potentially lingering offline thread when not online. */
224 if (!journal_file_is_offlining(f
) && f
->header
->state
!= STATE_ONLINE
)
225 return journal_file_set_offline_thread_join(f
);
227 /* Restart an in-flight offline thread and wait if needed, or join a lingering done one. */
228 restarted
= journal_file_set_offline_try_restart(f
);
229 if ((restarted
&& wait
) || !restarted
) {
230 r
= journal_file_set_offline_thread_join(f
);
238 /* Initiate a new offline. */
239 f
->offline_state
= OFFLINE_SYNCING
;
241 if (wait
) /* Without using a thread if waiting. */
242 journal_file_set_offline_internal(f
);
244 r
= pthread_create(&f
->offline_thread
, NULL
, journal_file_set_offline_thread
, f
);
246 f
->offline_state
= OFFLINE_JOINED
;
254 static int journal_file_set_online(JournalFile
*f
) {
262 if (!(f
->fd
>= 0 && f
->header
))
266 switch (f
->offline_state
) {
268 /* No offline thread, no need to wait. */
272 case OFFLINE_SYNCING
:
273 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_SYNCING
, OFFLINE_CANCEL
))
275 /* Canceled syncing prior to offlining, no need to wait. */
278 case OFFLINE_AGAIN_FROM_SYNCING
:
279 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_AGAIN_FROM_SYNCING
, OFFLINE_CANCEL
))
281 /* Canceled restart from syncing, no need to wait. */
284 case OFFLINE_AGAIN_FROM_OFFLINING
:
285 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_AGAIN_FROM_OFFLINING
, OFFLINE_CANCEL
))
287 /* Canceled restart from offlining, must wait for offlining to complete however. */
293 r
= journal_file_set_offline_thread_join(f
);
303 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
306 switch (f
->header
->state
) {
311 f
->header
->state
= STATE_ONLINE
;
320 bool journal_file_is_offlining(JournalFile
*f
) {
323 __sync_synchronize();
325 if (f
->offline_state
== OFFLINE_DONE
||
326 f
->offline_state
== OFFLINE_JOINED
)
332 JournalFile
* journal_file_close(JournalFile
*f
) {
336 /* Write the final tag */
337 if (f
->seal
&& f
->writable
) {
340 r
= journal_file_append_tag(f
);
342 log_error_errno(r
, "Failed to append tag when closing journal: %m");
346 if (f
->post_change_timer
) {
349 if (sd_event_source_get_enabled(f
->post_change_timer
, &enabled
) >= 0)
350 if (enabled
== SD_EVENT_ONESHOT
)
351 journal_file_post_change(f
);
353 (void) sd_event_source_set_enabled(f
->post_change_timer
, SD_EVENT_OFF
);
354 sd_event_source_unref(f
->post_change_timer
);
357 journal_file_set_offline(f
, true);
359 if (f
->mmap
&& f
->cache_fd
)
360 mmap_cache_free_fd(f
->mmap
, f
->cache_fd
);
362 if (f
->fd
>= 0 && f
->defrag_on_close
) {
364 /* Be friendly to btrfs: turn COW back on again now,
365 * and defragment the file. We won't write to the file
366 * ever again, hence remove all fragmentation, and
367 * reenable all the good bits COW usually provides
368 * (such as data checksumming). */
370 (void) chattr_fd(f
->fd
, 0, FS_NOCOW_FL
);
371 (void) btrfs_defrag_fd(f
->fd
);
378 mmap_cache_unref(f
->mmap
);
380 ordered_hashmap_free_free(f
->chain_cache
);
382 #if defined(HAVE_XZ) || defined(HAVE_LZ4)
383 free(f
->compress_buffer
);
388 munmap(f
->fss_file
, PAGE_ALIGN(f
->fss_file_size
));
390 free(f
->fsprg_state
);
395 gcry_md_close(f
->hmac
);
401 void journal_file_close_set(Set
*s
) {
406 while ((f
= set_steal_first(s
)))
407 (void) journal_file_close(f
);
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 fsync_directory_of_file(int fd
) {
448 _cleanup_free_
char *path
= NULL
, *dn
= NULL
;
449 _cleanup_close_
int dfd
= -1;
453 if (fstat(fd
, &st
) < 0)
456 if (!S_ISREG(st
.st_mode
))
459 r
= fd_get_path(fd
, &path
);
463 if (!path_is_absolute(path
))
466 dn
= dirname_malloc(path
);
470 dfd
= open(dn
, O_RDONLY
|O_CLOEXEC
|O_DIRECTORY
);
480 static int journal_file_refresh_header(JournalFile
*f
) {
487 r
= sd_id128_get_machine(&f
->header
->machine_id
);
491 r
= sd_id128_get_boot(&boot_id
);
495 if (sd_id128_equal(boot_id
, f
->header
->boot_id
))
496 f
->tail_entry_monotonic_valid
= true;
498 f
->header
->boot_id
= boot_id
;
500 r
= journal_file_set_online(f
);
502 /* Sync the online state to disk */
505 /* We likely just created a new file, also sync the directory this file is located in. */
506 (void) fsync_directory_of_file(f
->fd
);
511 static bool warn_wrong_flags(const JournalFile
*f
, bool compatible
) {
512 const uint32_t any
= compatible
? HEADER_COMPATIBLE_ANY
: HEADER_INCOMPATIBLE_ANY
,
513 supported
= compatible
? HEADER_COMPATIBLE_SUPPORTED
: HEADER_INCOMPATIBLE_SUPPORTED
;
514 const char *type
= compatible
? "compatible" : "incompatible";
517 flags
= le32toh(compatible
? f
->header
->compatible_flags
: f
->header
->incompatible_flags
);
519 if (flags
& ~supported
) {
521 log_debug("Journal file %s has unknown %s flags 0x%"PRIx32
,
522 f
->path
, type
, flags
& ~any
);
523 flags
= (flags
& any
) & ~supported
;
527 _cleanup_free_
char *t
= NULL
;
529 if (compatible
&& (flags
& HEADER_COMPATIBLE_SEALED
))
530 strv
[n
++] = "sealed";
531 if (!compatible
&& (flags
& HEADER_INCOMPATIBLE_COMPRESSED_XZ
))
532 strv
[n
++] = "xz-compressed";
533 if (!compatible
&& (flags
& HEADER_INCOMPATIBLE_COMPRESSED_LZ4
))
534 strv
[n
++] = "lz4-compressed";
536 assert(n
< ELEMENTSOF(strv
));
538 t
= strv_join((char**) strv
, ", ");
539 log_debug("Journal file %s uses %s %s %s disabled at compilation time.",
540 f
->path
, type
, n
> 1 ? "flags" : "flag", strnull(t
));
548 static int journal_file_verify_header(JournalFile
*f
) {
549 uint64_t arena_size
, header_size
;
554 if (memcmp(f
->header
->signature
, HEADER_SIGNATURE
, 8))
557 /* In both read and write mode we refuse to open files with incompatible
558 * flags we don't know. */
559 if (warn_wrong_flags(f
, false))
560 return -EPROTONOSUPPORT
;
562 /* When open for writing we refuse to open files with compatible flags, too. */
563 if (f
->writable
&& warn_wrong_flags(f
, true))
564 return -EPROTONOSUPPORT
;
566 if (f
->header
->state
>= _STATE_MAX
)
569 header_size
= le64toh(f
->header
->header_size
);
571 /* The first addition was n_data, so check that we are at least this large */
572 if (header_size
< HEADER_SIZE_MIN
)
575 if (JOURNAL_HEADER_SEALED(f
->header
) && !JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
578 arena_size
= le64toh(f
->header
->arena_size
);
580 if (UINT64_MAX
- header_size
< arena_size
|| header_size
+ arena_size
> (uint64_t) f
->last_stat
.st_size
)
583 if (le64toh(f
->header
->tail_object_offset
) > header_size
+ arena_size
)
586 if (!VALID64(le64toh(f
->header
->data_hash_table_offset
)) ||
587 !VALID64(le64toh(f
->header
->field_hash_table_offset
)) ||
588 !VALID64(le64toh(f
->header
->tail_object_offset
)) ||
589 !VALID64(le64toh(f
->header
->entry_array_offset
)))
593 sd_id128_t machine_id
;
597 r
= sd_id128_get_machine(&machine_id
);
601 if (!sd_id128_equal(machine_id
, f
->header
->machine_id
))
604 state
= f
->header
->state
;
606 if (state
== STATE_ARCHIVED
)
607 return -ESHUTDOWN
; /* Already archived */
608 else if (state
== STATE_ONLINE
) {
609 log_debug("Journal file %s is already online. Assuming unclean closing.", f
->path
);
611 } else if (state
!= STATE_OFFLINE
) {
612 log_debug("Journal file %s has unknown state %i.", f
->path
, state
);
616 if (f
->header
->field_hash_table_size
== 0 || f
->header
->data_hash_table_size
== 0)
619 /* Don't permit appending to files from the future. Because otherwise the realtime timestamps wouldn't
620 * be strictly ordered in the entries in the file anymore, and we can't have that since it breaks
622 if (le64toh(f
->header
->tail_entry_realtime
) > now(CLOCK_REALTIME
)) {
623 log_debug("Journal file %s is from the future, refusing to append new data to it that'd be older.", f
->path
);
628 f
->compress_xz
= JOURNAL_HEADER_COMPRESSED_XZ(f
->header
);
629 f
->compress_lz4
= JOURNAL_HEADER_COMPRESSED_LZ4(f
->header
);
631 f
->seal
= JOURNAL_HEADER_SEALED(f
->header
);
636 static int journal_file_fstat(JournalFile
*f
) {
640 if (fstat(f
->fd
, &f
->last_stat
) < 0)
643 f
->last_stat_usec
= now(CLOCK_MONOTONIC
);
645 /* Refuse appending to files that are already deleted */
646 if (f
->last_stat
.st_nlink
<= 0)
652 static int journal_file_allocate(JournalFile
*f
, uint64_t offset
, uint64_t size
) {
653 uint64_t old_size
, new_size
;
659 /* We assume that this file is not sparse, and we know that
660 * for sure, since we always call posix_fallocate()
663 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
667 le64toh(f
->header
->header_size
) +
668 le64toh(f
->header
->arena_size
);
670 new_size
= PAGE_ALIGN(offset
+ size
);
671 if (new_size
< le64toh(f
->header
->header_size
))
672 new_size
= le64toh(f
->header
->header_size
);
674 if (new_size
<= old_size
) {
676 /* We already pre-allocated enough space, but before
677 * we write to it, let's check with fstat() if the
678 * file got deleted, in order make sure we don't throw
679 * away the data immediately. Don't check fstat() for
680 * all writes though, but only once ever 10s. */
682 if (f
->last_stat_usec
+ LAST_STAT_REFRESH_USEC
> now(CLOCK_MONOTONIC
))
685 return journal_file_fstat(f
);
688 /* Allocate more space. */
690 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
693 if (new_size
> f
->metrics
.min_size
&& f
->metrics
.keep_free
> 0) {
696 if (fstatvfs(f
->fd
, &svfs
) >= 0) {
699 available
= LESS_BY((uint64_t) svfs
.f_bfree
* (uint64_t) svfs
.f_bsize
, f
->metrics
.keep_free
);
701 if (new_size
- old_size
> available
)
706 /* Increase by larger blocks at once */
707 new_size
= ((new_size
+FILE_SIZE_INCREASE
-1) / FILE_SIZE_INCREASE
) * FILE_SIZE_INCREASE
;
708 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
709 new_size
= f
->metrics
.max_size
;
711 /* Note that the glibc fallocate() fallback is very
712 inefficient, hence we try to minimize the allocation area
714 r
= posix_fallocate(f
->fd
, old_size
, new_size
- old_size
);
718 f
->header
->arena_size
= htole64(new_size
- le64toh(f
->header
->header_size
));
720 return journal_file_fstat(f
);
723 static unsigned type_to_context(ObjectType type
) {
724 /* One context for each type, plus one catch-all for the rest */
725 assert_cc(_OBJECT_TYPE_MAX
<= MMAP_CACHE_MAX_CONTEXTS
);
726 assert_cc(CONTEXT_HEADER
< MMAP_CACHE_MAX_CONTEXTS
);
727 return type
> OBJECT_UNUSED
&& type
< _OBJECT_TYPE_MAX
? type
: 0;
730 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
) {
739 /* Avoid SIGBUS on invalid accesses */
740 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
) {
741 /* Hmm, out of range? Let's refresh the fstat() data
742 * first, before we trust that check. */
744 r
= journal_file_fstat(f
);
748 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
)
749 return -EADDRNOTAVAIL
;
752 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
);
755 static uint64_t minimum_header_size(Object
*o
) {
757 static const uint64_t table
[] = {
758 [OBJECT_DATA
] = sizeof(DataObject
),
759 [OBJECT_FIELD
] = sizeof(FieldObject
),
760 [OBJECT_ENTRY
] = sizeof(EntryObject
),
761 [OBJECT_DATA_HASH_TABLE
] = sizeof(HashTableObject
),
762 [OBJECT_FIELD_HASH_TABLE
] = sizeof(HashTableObject
),
763 [OBJECT_ENTRY_ARRAY
] = sizeof(EntryArrayObject
),
764 [OBJECT_TAG
] = sizeof(TagObject
),
767 if (o
->object
.type
>= ELEMENTSOF(table
) || table
[o
->object
.type
] <= 0)
768 return sizeof(ObjectHeader
);
770 return table
[o
->object
.type
];
773 /* Lightweight object checks. We want this to be fast, so that we won't
774 * slowdown every journal_file_move_to_object() call too much. */
775 static int journal_file_check_object(JournalFile
*f
, uint64_t offset
, Object
*o
) {
779 switch (o
->object
.type
) {
782 if ((le64toh(o
->data
.entry_offset
) == 0) ^ (le64toh(o
->data
.n_entries
) == 0)) {
783 log_debug("Bad n_entries: %"PRIu64
": %"PRIu64
,
784 le64toh(o
->data
.n_entries
), offset
);
788 if (le64toh(o
->object
.size
) - offsetof(DataObject
, payload
) <= 0) {
789 log_debug("Bad object size (<= %zu): %"PRIu64
": %"PRIu64
,
790 offsetof(DataObject
, payload
),
791 le64toh(o
->object
.size
),
796 if (!VALID64(le64toh(o
->data
.next_hash_offset
)) ||
797 !VALID64(le64toh(o
->data
.next_field_offset
)) ||
798 !VALID64(le64toh(o
->data
.entry_offset
)) ||
799 !VALID64(le64toh(o
->data
.entry_array_offset
))) {
800 log_debug("Invalid offset, next_hash_offset="OFSfmt
", next_field_offset="OFSfmt
801 ", entry_offset="OFSfmt
", entry_array_offset="OFSfmt
": %"PRIu64
,
802 le64toh(o
->data
.next_hash_offset
),
803 le64toh(o
->data
.next_field_offset
),
804 le64toh(o
->data
.entry_offset
),
805 le64toh(o
->data
.entry_array_offset
),
814 if (le64toh(o
->object
.size
) - offsetof(FieldObject
, payload
) <= 0) {
816 "Bad field size (<= %zu): %"PRIu64
": %"PRIu64
,
817 offsetof(FieldObject
, payload
),
818 le64toh(o
->object
.size
),
823 if (!VALID64(le64toh(o
->field
.next_hash_offset
)) ||
824 !VALID64(le64toh(o
->field
.head_data_offset
))) {
826 "Invalid offset, next_hash_offset="OFSfmt
827 ", head_data_offset="OFSfmt
": %"PRIu64
,
828 le64toh(o
->field
.next_hash_offset
),
829 le64toh(o
->field
.head_data_offset
),
836 if ((le64toh(o
->object
.size
) - offsetof(EntryObject
, items
)) % sizeof(EntryItem
) != 0) {
838 "Bad entry size (<= %zu): %"PRIu64
": %"PRIu64
,
839 offsetof(EntryObject
, items
),
840 le64toh(o
->object
.size
),
845 if ((le64toh(o
->object
.size
) - offsetof(EntryObject
, items
)) / sizeof(EntryItem
) <= 0) {
847 "Invalid number items in entry: %"PRIu64
": %"PRIu64
,
848 (le64toh(o
->object
.size
) - offsetof(EntryObject
, items
)) / sizeof(EntryItem
),
853 if (le64toh(o
->entry
.seqnum
) <= 0) {
855 "Invalid entry seqnum: %"PRIx64
": %"PRIu64
,
856 le64toh(o
->entry
.seqnum
),
861 if (!VALID_REALTIME(le64toh(o
->entry
.realtime
))) {
863 "Invalid entry realtime timestamp: %"PRIu64
": %"PRIu64
,
864 le64toh(o
->entry
.realtime
),
869 if (!VALID_MONOTONIC(le64toh(o
->entry
.monotonic
))) {
871 "Invalid entry monotonic timestamp: %"PRIu64
": %"PRIu64
,
872 le64toh(o
->entry
.monotonic
),
879 case OBJECT_DATA_HASH_TABLE
:
880 case OBJECT_FIELD_HASH_TABLE
:
881 if ((le64toh(o
->object
.size
) - offsetof(HashTableObject
, items
)) % sizeof(HashItem
) != 0 ||
882 (le64toh(o
->object
.size
) - offsetof(HashTableObject
, items
)) / sizeof(HashItem
) <= 0) {
884 "Invalid %s hash table size: %"PRIu64
": %"PRIu64
,
885 o
->object
.type
== OBJECT_DATA_HASH_TABLE
? "data" : "field",
886 le64toh(o
->object
.size
),
893 case OBJECT_ENTRY_ARRAY
:
894 if ((le64toh(o
->object
.size
) - offsetof(EntryArrayObject
, items
)) % sizeof(le64_t
) != 0 ||
895 (le64toh(o
->object
.size
) - offsetof(EntryArrayObject
, items
)) / sizeof(le64_t
) <= 0) {
897 "Invalid object entry array size: %"PRIu64
": %"PRIu64
,
898 le64toh(o
->object
.size
),
903 if (!VALID64(le64toh(o
->entry_array
.next_entry_array_offset
))) {
905 "Invalid object entry array next_entry_array_offset: "OFSfmt
": %"PRIu64
,
906 le64toh(o
->entry_array
.next_entry_array_offset
),
914 if (le64toh(o
->object
.size
) != sizeof(TagObject
)) {
916 "Invalid object tag size: %"PRIu64
": %"PRIu64
,
917 le64toh(o
->object
.size
),
922 if (!VALID_EPOCH(le64toh(o
->tag
.epoch
))) {
924 "Invalid object tag epoch: %"PRIu64
": %"PRIu64
,
925 le64toh(o
->tag
.epoch
),
936 int journal_file_move_to_object(JournalFile
*f
, ObjectType type
, uint64_t offset
, Object
**ret
) {
946 /* Objects may only be located at multiple of 64 bit */
947 if (!VALID64(offset
)) {
948 log_debug("Attempt to move to object at non-64bit boundary: %" PRIu64
, offset
);
952 /* Object may not be located in the file header */
953 if (offset
< le64toh(f
->header
->header_size
)) {
954 log_debug("Attempt to move to object located in file header: %" PRIu64
, offset
);
958 r
= journal_file_move_to(f
, type
, false, offset
, sizeof(ObjectHeader
), &t
, &tsize
);
963 s
= le64toh(o
->object
.size
);
966 log_debug("Attempt to move to uninitialized object: %" PRIu64
, offset
);
969 if (s
< sizeof(ObjectHeader
)) {
970 log_debug("Attempt to move to overly short object: %" PRIu64
, offset
);
974 if (o
->object
.type
<= OBJECT_UNUSED
) {
975 log_debug("Attempt to move to object with invalid type: %" PRIu64
, offset
);
979 if (s
< minimum_header_size(o
)) {
980 log_debug("Attempt to move to truncated object: %" PRIu64
, offset
);
984 if (type
> OBJECT_UNUSED
&& o
->object
.type
!= type
) {
985 log_debug("Attempt to move to object of unexpected type: %" PRIu64
, offset
);
990 r
= journal_file_move_to(f
, type
, false, offset
, s
, &t
, NULL
);
997 r
= journal_file_check_object(f
, offset
, o
);
1005 static uint64_t journal_file_entry_seqnum(JournalFile
*f
, uint64_t *seqnum
) {
1011 r
= le64toh(f
->header
->tail_entry_seqnum
) + 1;
1014 /* If an external seqnum counter was passed, we update
1015 * both the local and the external one, and set it to
1016 * the maximum of both */
1018 if (*seqnum
+ 1 > r
)
1024 f
->header
->tail_entry_seqnum
= htole64(r
);
1026 if (f
->header
->head_entry_seqnum
== 0)
1027 f
->header
->head_entry_seqnum
= htole64(r
);
1032 int journal_file_append_object(JournalFile
*f
, ObjectType type
, uint64_t size
, Object
**ret
, uint64_t *offset
) {
1040 assert(type
> OBJECT_UNUSED
&& type
< _OBJECT_TYPE_MAX
);
1041 assert(size
>= sizeof(ObjectHeader
));
1045 r
= journal_file_set_online(f
);
1049 p
= le64toh(f
->header
->tail_object_offset
);
1051 p
= le64toh(f
->header
->header_size
);
1053 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &tail
);
1057 p
+= ALIGN64(le64toh(tail
->object
.size
));
1060 r
= journal_file_allocate(f
, p
, size
);
1064 r
= journal_file_move_to(f
, type
, false, p
, size
, &t
, NULL
);
1071 o
->object
.type
= type
;
1072 o
->object
.size
= htole64(size
);
1074 f
->header
->tail_object_offset
= htole64(p
);
1075 f
->header
->n_objects
= htole64(le64toh(f
->header
->n_objects
) + 1);
1083 static int journal_file_setup_data_hash_table(JournalFile
*f
) {
1091 /* We estimate that we need 1 hash table entry per 768 bytes
1092 of journal file and we want to make sure we never get
1093 beyond 75% fill level. Calculate the hash table size for
1094 the maximum file size based on these metrics. */
1096 s
= (f
->metrics
.max_size
* 4 / 768 / 3) * sizeof(HashItem
);
1097 if (s
< DEFAULT_DATA_HASH_TABLE_SIZE
)
1098 s
= DEFAULT_DATA_HASH_TABLE_SIZE
;
1100 log_debug("Reserving %"PRIu64
" entries in hash table.", s
/ sizeof(HashItem
));
1102 r
= journal_file_append_object(f
,
1103 OBJECT_DATA_HASH_TABLE
,
1104 offsetof(Object
, hash_table
.items
) + s
,
1109 memzero(o
->hash_table
.items
, s
);
1111 f
->header
->data_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
1112 f
->header
->data_hash_table_size
= htole64(s
);
1117 static int journal_file_setup_field_hash_table(JournalFile
*f
) {
1125 /* We use a fixed size hash table for the fields as this
1126 * number should grow very slowly only */
1128 s
= DEFAULT_FIELD_HASH_TABLE_SIZE
;
1129 r
= journal_file_append_object(f
,
1130 OBJECT_FIELD_HASH_TABLE
,
1131 offsetof(Object
, hash_table
.items
) + s
,
1136 memzero(o
->hash_table
.items
, s
);
1138 f
->header
->field_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
1139 f
->header
->field_hash_table_size
= htole64(s
);
1144 int journal_file_map_data_hash_table(JournalFile
*f
) {
1152 if (f
->data_hash_table
)
1155 p
= le64toh(f
->header
->data_hash_table_offset
);
1156 s
= le64toh(f
->header
->data_hash_table_size
);
1158 r
= journal_file_move_to(f
,
1159 OBJECT_DATA_HASH_TABLE
,
1166 f
->data_hash_table
= t
;
1170 int journal_file_map_field_hash_table(JournalFile
*f
) {
1178 if (f
->field_hash_table
)
1181 p
= le64toh(f
->header
->field_hash_table_offset
);
1182 s
= le64toh(f
->header
->field_hash_table_size
);
1184 r
= journal_file_move_to(f
,
1185 OBJECT_FIELD_HASH_TABLE
,
1192 f
->field_hash_table
= t
;
1196 static int journal_file_link_field(
1207 assert(f
->field_hash_table
);
1211 if (o
->object
.type
!= OBJECT_FIELD
)
1214 m
= le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
);
1218 /* This might alter the window we are looking at */
1219 o
->field
.next_hash_offset
= o
->field
.head_data_offset
= 0;
1222 p
= le64toh(f
->field_hash_table
[h
].tail_hash_offset
);
1224 f
->field_hash_table
[h
].head_hash_offset
= htole64(offset
);
1226 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1230 o
->field
.next_hash_offset
= htole64(offset
);
1233 f
->field_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1235 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
1236 f
->header
->n_fields
= htole64(le64toh(f
->header
->n_fields
) + 1);
1241 static int journal_file_link_data(
1252 assert(f
->data_hash_table
);
1256 if (o
->object
.type
!= OBJECT_DATA
)
1259 m
= le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
);
1263 /* This might alter the window we are looking at */
1264 o
->data
.next_hash_offset
= o
->data
.next_field_offset
= 0;
1265 o
->data
.entry_offset
= o
->data
.entry_array_offset
= 0;
1266 o
->data
.n_entries
= 0;
1269 p
= le64toh(f
->data_hash_table
[h
].tail_hash_offset
);
1271 /* Only entry in the hash table is easy */
1272 f
->data_hash_table
[h
].head_hash_offset
= htole64(offset
);
1274 /* Move back to the previous data object, to patch in
1277 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1281 o
->data
.next_hash_offset
= htole64(offset
);
1284 f
->data_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1286 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
1287 f
->header
->n_data
= htole64(le64toh(f
->header
->n_data
) + 1);
1292 int journal_file_find_field_object_with_hash(
1294 const void *field
, uint64_t size
, uint64_t hash
,
1295 Object
**ret
, uint64_t *offset
) {
1297 uint64_t p
, osize
, h
, m
;
1302 assert(field
&& size
> 0);
1304 /* If the field hash table is empty, we can't find anything */
1305 if (le64toh(f
->header
->field_hash_table_size
) <= 0)
1308 /* Map the field hash table, if it isn't mapped yet. */
1309 r
= journal_file_map_field_hash_table(f
);
1313 osize
= offsetof(Object
, field
.payload
) + size
;
1315 m
= le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
);
1320 p
= le64toh(f
->field_hash_table
[h
].head_hash_offset
);
1325 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1329 if (le64toh(o
->field
.hash
) == hash
&&
1330 le64toh(o
->object
.size
) == osize
&&
1331 memcmp(o
->field
.payload
, field
, size
) == 0) {
1341 p
= le64toh(o
->field
.next_hash_offset
);
1347 int journal_file_find_field_object(
1349 const void *field
, uint64_t size
,
1350 Object
**ret
, uint64_t *offset
) {
1355 assert(field
&& size
> 0);
1357 hash
= hash64(field
, size
);
1359 return journal_file_find_field_object_with_hash(f
,
1364 int journal_file_find_data_object_with_hash(
1366 const void *data
, uint64_t size
, uint64_t hash
,
1367 Object
**ret
, uint64_t *offset
) {
1369 uint64_t p
, osize
, h
, m
;
1374 assert(data
|| size
== 0);
1376 /* If there's no data hash table, then there's no entry. */
1377 if (le64toh(f
->header
->data_hash_table_size
) <= 0)
1380 /* Map the data hash table, if it isn't mapped yet. */
1381 r
= journal_file_map_data_hash_table(f
);
1385 osize
= offsetof(Object
, data
.payload
) + size
;
1387 m
= le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
);
1392 p
= le64toh(f
->data_hash_table
[h
].head_hash_offset
);
1397 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1401 if (le64toh(o
->data
.hash
) != hash
)
1404 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
) {
1405 #if defined(HAVE_XZ) || defined(HAVE_LZ4)
1409 l
= le64toh(o
->object
.size
);
1410 if (l
<= offsetof(Object
, data
.payload
))
1413 l
-= offsetof(Object
, data
.payload
);
1415 r
= decompress_blob(o
->object
.flags
& OBJECT_COMPRESSION_MASK
,
1416 o
->data
.payload
, l
, &f
->compress_buffer
, &f
->compress_buffer_size
, &rsize
, 0);
1420 if (rsize
== size
&&
1421 memcmp(f
->compress_buffer
, data
, size
) == 0) {
1432 return -EPROTONOSUPPORT
;
1434 } else if (le64toh(o
->object
.size
) == osize
&&
1435 memcmp(o
->data
.payload
, data
, size
) == 0) {
1447 p
= le64toh(o
->data
.next_hash_offset
);
1453 int journal_file_find_data_object(
1455 const void *data
, uint64_t size
,
1456 Object
**ret
, uint64_t *offset
) {
1461 assert(data
|| size
== 0);
1463 hash
= hash64(data
, size
);
1465 return journal_file_find_data_object_with_hash(f
,
1470 static int journal_file_append_field(
1472 const void *field
, uint64_t size
,
1473 Object
**ret
, uint64_t *offset
) {
1481 assert(field
&& size
> 0);
1483 hash
= hash64(field
, size
);
1485 r
= journal_file_find_field_object_with_hash(f
, field
, size
, hash
, &o
, &p
);
1499 osize
= offsetof(Object
, field
.payload
) + size
;
1500 r
= journal_file_append_object(f
, OBJECT_FIELD
, osize
, &o
, &p
);
1504 o
->field
.hash
= htole64(hash
);
1505 memcpy(o
->field
.payload
, field
, size
);
1507 r
= journal_file_link_field(f
, o
, p
, hash
);
1511 /* The linking might have altered the window, so let's
1512 * refresh our pointer */
1513 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1518 r
= journal_file_hmac_put_object(f
, OBJECT_FIELD
, o
, p
);
1532 static int journal_file_append_data(
1534 const void *data
, uint64_t size
,
1535 Object
**ret
, uint64_t *offset
) {
1540 int r
, compression
= 0;
1544 assert(data
|| size
== 0);
1546 hash
= hash64(data
, size
);
1548 r
= journal_file_find_data_object_with_hash(f
, data
, size
, hash
, &o
, &p
);
1562 osize
= offsetof(Object
, data
.payload
) + size
;
1563 r
= journal_file_append_object(f
, OBJECT_DATA
, osize
, &o
, &p
);
1567 o
->data
.hash
= htole64(hash
);
1569 #if defined(HAVE_XZ) || defined(HAVE_LZ4)
1570 if (JOURNAL_FILE_COMPRESS(f
) && size
>= COMPRESSION_SIZE_THRESHOLD
) {
1573 compression
= compress_blob(data
, size
, o
->data
.payload
, size
- 1, &rsize
);
1575 if (compression
>= 0) {
1576 o
->object
.size
= htole64(offsetof(Object
, data
.payload
) + rsize
);
1577 o
->object
.flags
|= compression
;
1579 log_debug("Compressed data object %"PRIu64
" -> %zu using %s",
1580 size
, rsize
, object_compressed_to_string(compression
));
1582 /* Compression didn't work, we don't really care why, let's continue without compression */
1587 if (compression
== 0)
1588 memcpy_safe(o
->data
.payload
, data
, size
);
1590 r
= journal_file_link_data(f
, o
, p
, hash
);
1595 r
= journal_file_hmac_put_object(f
, OBJECT_DATA
, o
, p
);
1600 /* The linking might have altered the window, so let's
1601 * refresh our pointer */
1602 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1609 eq
= memchr(data
, '=', size
);
1610 if (eq
&& eq
> data
) {
1614 /* Create field object ... */
1615 r
= journal_file_append_field(f
, data
, (uint8_t*) eq
- (uint8_t*) data
, &fo
, &fp
);
1619 /* ... and link it in. */
1620 o
->data
.next_field_offset
= fo
->field
.head_data_offset
;
1621 fo
->field
.head_data_offset
= le64toh(p
);
1633 uint64_t journal_file_entry_n_items(Object
*o
) {
1636 if (o
->object
.type
!= OBJECT_ENTRY
)
1639 return (le64toh(o
->object
.size
) - offsetof(Object
, entry
.items
)) / sizeof(EntryItem
);
1642 uint64_t journal_file_entry_array_n_items(Object
*o
) {
1645 if (o
->object
.type
!= OBJECT_ENTRY_ARRAY
)
1648 return (le64toh(o
->object
.size
) - offsetof(Object
, entry_array
.items
)) / sizeof(uint64_t);
1651 uint64_t journal_file_hash_table_n_items(Object
*o
) {
1654 if (o
->object
.type
!= OBJECT_DATA_HASH_TABLE
&&
1655 o
->object
.type
!= OBJECT_FIELD_HASH_TABLE
)
1658 return (le64toh(o
->object
.size
) - offsetof(Object
, hash_table
.items
)) / sizeof(HashItem
);
1661 static int link_entry_into_array(JournalFile
*f
,
1666 uint64_t n
= 0, ap
= 0, q
, i
, a
, hidx
;
1675 a
= le64toh(*first
);
1676 i
= hidx
= le64toh(*idx
);
1679 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
1683 n
= journal_file_entry_array_n_items(o
);
1685 o
->entry_array
.items
[i
] = htole64(p
);
1686 *idx
= htole64(hidx
+ 1);
1692 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
1703 r
= journal_file_append_object(f
, OBJECT_ENTRY_ARRAY
,
1704 offsetof(Object
, entry_array
.items
) + n
* sizeof(uint64_t),
1710 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY_ARRAY
, o
, q
);
1715 o
->entry_array
.items
[i
] = htole64(p
);
1718 *first
= htole64(q
);
1720 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, ap
, &o
);
1724 o
->entry_array
.next_entry_array_offset
= htole64(q
);
1727 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
1728 f
->header
->n_entry_arrays
= htole64(le64toh(f
->header
->n_entry_arrays
) + 1);
1730 *idx
= htole64(hidx
+ 1);
1735 static int link_entry_into_array_plus_one(JournalFile
*f
,
1750 *extra
= htole64(p
);
1754 i
= htole64(le64toh(*idx
) - 1);
1755 r
= link_entry_into_array(f
, first
, &i
, p
);
1760 *idx
= htole64(le64toh(*idx
) + 1);
1764 static int journal_file_link_entry_item(JournalFile
*f
, Object
*o
, uint64_t offset
, uint64_t i
) {
1771 p
= le64toh(o
->entry
.items
[i
].object_offset
);
1775 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1779 return link_entry_into_array_plus_one(f
,
1780 &o
->data
.entry_offset
,
1781 &o
->data
.entry_array_offset
,
1786 static int journal_file_link_entry(JournalFile
*f
, Object
*o
, uint64_t offset
) {
1795 if (o
->object
.type
!= OBJECT_ENTRY
)
1798 __sync_synchronize();
1800 /* Link up the entry itself */
1801 r
= link_entry_into_array(f
,
1802 &f
->header
->entry_array_offset
,
1803 &f
->header
->n_entries
,
1808 /* log_debug("=> %s seqnr=%"PRIu64" n_entries=%"PRIu64, f->path, o->entry.seqnum, f->header->n_entries); */
1810 if (f
->header
->head_entry_realtime
== 0)
1811 f
->header
->head_entry_realtime
= o
->entry
.realtime
;
1813 f
->header
->tail_entry_realtime
= o
->entry
.realtime
;
1814 f
->header
->tail_entry_monotonic
= o
->entry
.monotonic
;
1816 f
->tail_entry_monotonic_valid
= true;
1818 /* Link up the items */
1819 n
= journal_file_entry_n_items(o
);
1820 for (i
= 0; i
< n
; i
++) {
1821 r
= journal_file_link_entry_item(f
, o
, offset
, i
);
1829 static int journal_file_append_entry_internal(
1831 const dual_timestamp
*ts
,
1833 const EntryItem items
[], unsigned n_items
,
1835 Object
**ret
, uint64_t *offset
) {
1843 assert(items
|| n_items
== 0);
1846 osize
= offsetof(Object
, entry
.items
) + (n_items
* sizeof(EntryItem
));
1848 r
= journal_file_append_object(f
, OBJECT_ENTRY
, osize
, &o
, &np
);
1852 o
->entry
.seqnum
= htole64(journal_file_entry_seqnum(f
, seqnum
));
1853 memcpy_safe(o
->entry
.items
, items
, n_items
* sizeof(EntryItem
));
1854 o
->entry
.realtime
= htole64(ts
->realtime
);
1855 o
->entry
.monotonic
= htole64(ts
->monotonic
);
1856 o
->entry
.xor_hash
= htole64(xor_hash
);
1857 o
->entry
.boot_id
= f
->header
->boot_id
;
1860 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY
, o
, np
);
1865 r
= journal_file_link_entry(f
, o
, np
);
1878 void journal_file_post_change(JournalFile
*f
) {
1881 /* inotify() does not receive IN_MODIFY events from file
1882 * accesses done via mmap(). After each access we hence
1883 * trigger IN_MODIFY by truncating the journal file to its
1884 * current size which triggers IN_MODIFY. */
1886 __sync_synchronize();
1888 if (ftruncate(f
->fd
, f
->last_stat
.st_size
) < 0)
1889 log_debug_errno(errno
, "Failed to truncate file to its own size: %m");
1892 static int post_change_thunk(sd_event_source
*timer
, uint64_t usec
, void *userdata
) {
1895 journal_file_post_change(userdata
);
1900 static void schedule_post_change(JournalFile
*f
) {
1901 sd_event_source
*timer
;
1906 assert(f
->post_change_timer
);
1908 timer
= f
->post_change_timer
;
1910 r
= sd_event_source_get_enabled(timer
, &enabled
);
1912 log_debug_errno(r
, "Failed to get ftruncate timer state: %m");
1916 if (enabled
== SD_EVENT_ONESHOT
)
1919 r
= sd_event_now(sd_event_source_get_event(timer
), CLOCK_MONOTONIC
, &now
);
1921 log_debug_errno(r
, "Failed to get clock's now for scheduling ftruncate: %m");
1925 r
= sd_event_source_set_time(timer
, now
+f
->post_change_timer_period
);
1927 log_debug_errno(r
, "Failed to set time for scheduling ftruncate: %m");
1931 r
= sd_event_source_set_enabled(timer
, SD_EVENT_ONESHOT
);
1933 log_debug_errno(r
, "Failed to enable scheduled ftruncate: %m");
1940 /* On failure, let's simply post the change immediately. */
1941 journal_file_post_change(f
);
1944 /* Enable coalesced change posting in a timer on the provided sd_event instance */
1945 int journal_file_enable_post_change_timer(JournalFile
*f
, sd_event
*e
, usec_t t
) {
1946 _cleanup_(sd_event_source_unrefp
) sd_event_source
*timer
= NULL
;
1950 assert_return(!f
->post_change_timer
, -EINVAL
);
1954 r
= sd_event_add_time(e
, &timer
, CLOCK_MONOTONIC
, 0, 0, post_change_thunk
, f
);
1958 r
= sd_event_source_set_enabled(timer
, SD_EVENT_OFF
);
1962 f
->post_change_timer
= timer
;
1964 f
->post_change_timer_period
= t
;
1969 static int entry_item_cmp(const void *_a
, const void *_b
) {
1970 const EntryItem
*a
= _a
, *b
= _b
;
1972 if (le64toh(a
->object_offset
) < le64toh(b
->object_offset
))
1974 if (le64toh(a
->object_offset
) > le64toh(b
->object_offset
))
1979 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
) {
1983 uint64_t xor_hash
= 0;
1984 struct dual_timestamp _ts
;
1988 assert(iovec
|| n_iovec
== 0);
1991 dual_timestamp_get(&_ts
);
1996 r
= journal_file_maybe_append_tag(f
, ts
->realtime
);
2001 /* alloca() can't take 0, hence let's allocate at least one */
2002 items
= alloca(sizeof(EntryItem
) * MAX(1u, n_iovec
));
2004 for (i
= 0; i
< n_iovec
; i
++) {
2008 r
= journal_file_append_data(f
, iovec
[i
].iov_base
, iovec
[i
].iov_len
, &o
, &p
);
2012 xor_hash
^= le64toh(o
->data
.hash
);
2013 items
[i
].object_offset
= htole64(p
);
2014 items
[i
].hash
= o
->data
.hash
;
2017 /* Order by the position on disk, in order to improve seek
2018 * times for rotating media. */
2019 qsort_safe(items
, n_iovec
, sizeof(EntryItem
), entry_item_cmp
);
2021 r
= journal_file_append_entry_internal(f
, ts
, xor_hash
, items
, n_iovec
, seqnum
, ret
, offset
);
2023 /* If the memory mapping triggered a SIGBUS then we return an
2024 * IO error and ignore the error code passed down to us, since
2025 * it is very likely just an effect of a nullified replacement
2028 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
2031 if (f
->post_change_timer
)
2032 schedule_post_change(f
);
2034 journal_file_post_change(f
);
2039 typedef struct ChainCacheItem
{
2040 uint64_t first
; /* the array at the beginning of the chain */
2041 uint64_t array
; /* the cached array */
2042 uint64_t begin
; /* the first item in the cached array */
2043 uint64_t total
; /* the total number of items in all arrays before this one in the chain */
2044 uint64_t last_index
; /* the last index we looked at, to optimize locality when bisecting */
2047 static void chain_cache_put(
2054 uint64_t last_index
) {
2057 /* If the chain item to cache for this chain is the
2058 * first one it's not worth caching anything */
2062 if (ordered_hashmap_size(h
) >= CHAIN_CACHE_MAX
) {
2063 ci
= ordered_hashmap_steal_first(h
);
2066 ci
= new(ChainCacheItem
, 1);
2073 if (ordered_hashmap_put(h
, &ci
->first
, ci
) < 0) {
2078 assert(ci
->first
== first
);
2083 ci
->last_index
= last_index
;
2086 static int generic_array_get(
2090 Object
**ret
, uint64_t *offset
) {
2093 uint64_t p
= 0, a
, t
= 0;
2101 /* Try the chain cache first */
2102 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
2103 if (ci
&& i
> ci
->total
) {
2112 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
2116 k
= journal_file_entry_array_n_items(o
);
2118 p
= le64toh(o
->entry_array
.items
[i
]);
2124 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
2130 /* Let's cache this item for the next invocation */
2131 chain_cache_put(f
->chain_cache
, ci
, first
, a
, le64toh(o
->entry_array
.items
[0]), t
, i
);
2133 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2146 static int generic_array_get_plus_one(
2151 Object
**ret
, uint64_t *offset
) {
2160 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, &o
);
2173 return generic_array_get(f
, first
, i
-1, ret
, offset
);
2182 static int generic_array_bisect(
2187 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
2188 direction_t direction
,
2193 uint64_t a
, p
, t
= 0, i
= 0, last_p
= 0, last_index
= (uint64_t) -1;
2194 bool subtract_one
= false;
2195 Object
*o
, *array
= NULL
;
2200 assert(test_object
);
2202 /* Start with the first array in the chain */
2205 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
2206 if (ci
&& n
> ci
->total
) {
2207 /* Ah, we have iterated this bisection array chain
2208 * previously! Let's see if we can skip ahead in the
2209 * chain, as far as the last time. But we can't jump
2210 * backwards in the chain, so let's check that
2213 r
= test_object(f
, ci
->begin
, needle
);
2217 if (r
== TEST_LEFT
) {
2218 /* OK, what we are looking for is right of the
2219 * begin of this EntryArray, so let's jump
2220 * straight to previously cached array in the
2226 last_index
= ci
->last_index
;
2231 uint64_t left
, right
, k
, lp
;
2233 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &array
);
2237 k
= journal_file_entry_array_n_items(array
);
2243 lp
= p
= le64toh(array
->entry_array
.items
[i
]);
2247 r
= test_object(f
, p
, needle
);
2248 if (r
== -EBADMSG
) {
2249 log_debug_errno(r
, "Encountered invalid entry while bisecting, cutting algorithm short. (1)");
2256 if (r
== TEST_FOUND
)
2257 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2259 if (r
== TEST_RIGHT
) {
2263 if (last_index
!= (uint64_t) -1) {
2264 assert(last_index
<= right
);
2266 /* If we cached the last index we
2267 * looked at, let's try to not to jump
2268 * too wildly around and see if we can
2269 * limit the range to look at early to
2270 * the immediate neighbors of the last
2271 * index we looked at. */
2273 if (last_index
> 0) {
2274 uint64_t x
= last_index
- 1;
2276 p
= le64toh(array
->entry_array
.items
[x
]);
2280 r
= test_object(f
, p
, needle
);
2284 if (r
== TEST_FOUND
)
2285 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2287 if (r
== TEST_RIGHT
)
2293 if (last_index
< right
) {
2294 uint64_t y
= last_index
+ 1;
2296 p
= le64toh(array
->entry_array
.items
[y
]);
2300 r
= test_object(f
, p
, needle
);
2304 if (r
== TEST_FOUND
)
2305 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2307 if (r
== TEST_RIGHT
)
2315 if (left
== right
) {
2316 if (direction
== DIRECTION_UP
)
2317 subtract_one
= true;
2323 assert(left
< right
);
2324 i
= (left
+ right
) / 2;
2326 p
= le64toh(array
->entry_array
.items
[i
]);
2330 r
= test_object(f
, p
, needle
);
2331 if (r
== -EBADMSG
) {
2332 log_debug_errno(r
, "Encountered invalid entry while bisecting, cutting algorithm short. (2)");
2339 if (r
== TEST_FOUND
)
2340 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2342 if (r
== TEST_RIGHT
)
2350 if (direction
== DIRECTION_UP
) {
2352 subtract_one
= true;
2363 last_index
= (uint64_t) -1;
2364 a
= le64toh(array
->entry_array
.next_entry_array_offset
);
2370 if (subtract_one
&& t
== 0 && i
== 0)
2373 /* Let's cache this item for the next invocation */
2374 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
);
2376 if (subtract_one
&& i
== 0)
2378 else if (subtract_one
)
2379 p
= le64toh(array
->entry_array
.items
[i
-1]);
2381 p
= le64toh(array
->entry_array
.items
[i
]);
2383 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2394 *idx
= t
+ i
+ (subtract_one
? -1 : 0);
2399 static int generic_array_bisect_plus_one(
2405 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
2406 direction_t direction
,
2412 bool step_back
= false;
2416 assert(test_object
);
2421 /* This bisects the array in object 'first', but first checks
2423 r
= test_object(f
, extra
, needle
);
2427 if (r
== TEST_FOUND
)
2428 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2430 /* if we are looking with DIRECTION_UP then we need to first
2431 see if in the actual array there is a matching entry, and
2432 return the last one of that. But if there isn't any we need
2433 to return this one. Hence remember this, and return it
2436 step_back
= direction
== DIRECTION_UP
;
2438 if (r
== TEST_RIGHT
) {
2439 if (direction
== DIRECTION_DOWN
)
2445 r
= generic_array_bisect(f
, first
, n
-1, needle
, test_object
, direction
, ret
, offset
, idx
);
2447 if (r
== 0 && step_back
)
2456 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, &o
);
2472 _pure_
static int test_object_offset(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2478 else if (p
< needle
)
2484 static int test_object_seqnum(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2491 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2495 if (le64toh(o
->entry
.seqnum
) == needle
)
2497 else if (le64toh(o
->entry
.seqnum
) < needle
)
2503 int journal_file_move_to_entry_by_seqnum(
2506 direction_t direction
,
2512 return generic_array_bisect(f
,
2513 le64toh(f
->header
->entry_array_offset
),
2514 le64toh(f
->header
->n_entries
),
2521 static int test_object_realtime(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2528 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2532 if (le64toh(o
->entry
.realtime
) == needle
)
2534 else if (le64toh(o
->entry
.realtime
) < needle
)
2540 int journal_file_move_to_entry_by_realtime(
2543 direction_t direction
,
2549 return generic_array_bisect(f
,
2550 le64toh(f
->header
->entry_array_offset
),
2551 le64toh(f
->header
->n_entries
),
2553 test_object_realtime
,
2558 static int test_object_monotonic(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2565 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2569 if (le64toh(o
->entry
.monotonic
) == needle
)
2571 else if (le64toh(o
->entry
.monotonic
) < needle
)
2577 static int find_data_object_by_boot_id(
2583 char t
[sizeof("_BOOT_ID=")-1 + 32 + 1] = "_BOOT_ID=";
2585 sd_id128_to_string(boot_id
, t
+ 9);
2586 return journal_file_find_data_object(f
, t
, sizeof(t
) - 1, o
, b
);
2589 int journal_file_move_to_entry_by_monotonic(
2593 direction_t direction
,
2602 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, NULL
);
2608 return generic_array_bisect_plus_one(f
,
2609 le64toh(o
->data
.entry_offset
),
2610 le64toh(o
->data
.entry_array_offset
),
2611 le64toh(o
->data
.n_entries
),
2613 test_object_monotonic
,
2618 void journal_file_reset_location(JournalFile
*f
) {
2619 f
->location_type
= LOCATION_HEAD
;
2620 f
->current_offset
= 0;
2621 f
->current_seqnum
= 0;
2622 f
->current_realtime
= 0;
2623 f
->current_monotonic
= 0;
2624 zero(f
->current_boot_id
);
2625 f
->current_xor_hash
= 0;
2628 void journal_file_save_location(JournalFile
*f
, Object
*o
, uint64_t offset
) {
2629 f
->location_type
= LOCATION_SEEK
;
2630 f
->current_offset
= offset
;
2631 f
->current_seqnum
= le64toh(o
->entry
.seqnum
);
2632 f
->current_realtime
= le64toh(o
->entry
.realtime
);
2633 f
->current_monotonic
= le64toh(o
->entry
.monotonic
);
2634 f
->current_boot_id
= o
->entry
.boot_id
;
2635 f
->current_xor_hash
= le64toh(o
->entry
.xor_hash
);
2638 int journal_file_compare_locations(JournalFile
*af
, JournalFile
*bf
) {
2643 assert(af
->location_type
== LOCATION_SEEK
);
2644 assert(bf
->location_type
== LOCATION_SEEK
);
2646 /* If contents and timestamps match, these entries are
2647 * identical, even if the seqnum does not match */
2648 if (sd_id128_equal(af
->current_boot_id
, bf
->current_boot_id
) &&
2649 af
->current_monotonic
== bf
->current_monotonic
&&
2650 af
->current_realtime
== bf
->current_realtime
&&
2651 af
->current_xor_hash
== bf
->current_xor_hash
)
2654 if (sd_id128_equal(af
->header
->seqnum_id
, bf
->header
->seqnum_id
)) {
2656 /* If this is from the same seqnum source, compare
2658 if (af
->current_seqnum
< bf
->current_seqnum
)
2660 if (af
->current_seqnum
> bf
->current_seqnum
)
2663 /* Wow! This is weird, different data but the same
2664 * seqnums? Something is borked, but let's make the
2665 * best of it and compare by time. */
2668 if (sd_id128_equal(af
->current_boot_id
, bf
->current_boot_id
)) {
2670 /* If the boot id matches, compare monotonic time */
2671 if (af
->current_monotonic
< bf
->current_monotonic
)
2673 if (af
->current_monotonic
> bf
->current_monotonic
)
2677 /* Otherwise, compare UTC time */
2678 if (af
->current_realtime
< bf
->current_realtime
)
2680 if (af
->current_realtime
> bf
->current_realtime
)
2683 /* Finally, compare by contents */
2684 if (af
->current_xor_hash
< bf
->current_xor_hash
)
2686 if (af
->current_xor_hash
> bf
->current_xor_hash
)
2692 static int bump_array_index(uint64_t *i
, direction_t direction
, uint64_t n
) {
2694 /* Increase or decrease the specified index, in the right direction. */
2696 if (direction
== DIRECTION_DOWN
) {
2711 static bool check_properly_ordered(uint64_t new_offset
, uint64_t old_offset
, direction_t direction
) {
2713 /* Consider it an error if any of the two offsets is uninitialized */
2714 if (old_offset
== 0 || new_offset
== 0)
2717 /* If we go down, the new offset must be larger than the old one. */
2718 return direction
== DIRECTION_DOWN
?
2719 new_offset
> old_offset
:
2720 new_offset
< old_offset
;
2723 int journal_file_next_entry(
2726 direction_t direction
,
2727 Object
**ret
, uint64_t *offset
) {
2735 n
= le64toh(f
->header
->n_entries
);
2740 i
= direction
== DIRECTION_DOWN
? 0 : n
- 1;
2742 r
= generic_array_bisect(f
,
2743 le64toh(f
->header
->entry_array_offset
),
2744 le64toh(f
->header
->n_entries
),
2753 r
= bump_array_index(&i
, direction
, n
);
2758 /* And jump to it */
2760 r
= generic_array_get(f
,
2761 le64toh(f
->header
->entry_array_offset
),
2769 /* OK, so this entry is borked. Most likely some entry didn't get synced to disk properly, let's see if
2770 * the next one might work for us instead. */
2771 log_debug_errno(r
, "Entry item %" PRIu64
" is bad, skipping over it.", i
);
2773 r
= bump_array_index(&i
, direction
, n
);
2778 /* Ensure our array is properly ordered. */
2779 if (p
> 0 && !check_properly_ordered(ofs
, p
, direction
)) {
2780 log_debug("%s: entry array not properly ordered at entry %" PRIu64
, f
->path
, i
);
2790 int journal_file_next_entry_for_data(
2792 Object
*o
, uint64_t p
,
2793 uint64_t data_offset
,
2794 direction_t direction
,
2795 Object
**ret
, uint64_t *offset
) {
2802 assert(p
> 0 || !o
);
2804 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2808 n
= le64toh(d
->data
.n_entries
);
2813 i
= direction
== DIRECTION_DOWN
? 0 : n
- 1;
2815 if (o
->object
.type
!= OBJECT_ENTRY
)
2818 r
= generic_array_bisect_plus_one(f
,
2819 le64toh(d
->data
.entry_offset
),
2820 le64toh(d
->data
.entry_array_offset
),
2821 le64toh(d
->data
.n_entries
),
2831 r
= bump_array_index(&i
, direction
, n
);
2837 r
= generic_array_get_plus_one(f
,
2838 le64toh(d
->data
.entry_offset
),
2839 le64toh(d
->data
.entry_array_offset
),
2847 log_debug_errno(r
, "Data entry item %" PRIu64
" is bad, skipping over it.", i
);
2849 r
= bump_array_index(&i
, direction
, n
);
2854 /* Ensure our array is properly ordered. */
2855 if (p
> 0 && check_properly_ordered(ofs
, p
, direction
)) {
2856 log_debug("%s data entry array not properly ordered at entry %" PRIu64
, f
->path
, i
);
2866 int journal_file_move_to_entry_by_offset_for_data(
2868 uint64_t data_offset
,
2870 direction_t direction
,
2871 Object
**ret
, uint64_t *offset
) {
2878 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2882 return generic_array_bisect_plus_one(f
,
2883 le64toh(d
->data
.entry_offset
),
2884 le64toh(d
->data
.entry_array_offset
),
2885 le64toh(d
->data
.n_entries
),
2892 int journal_file_move_to_entry_by_monotonic_for_data(
2894 uint64_t data_offset
,
2897 direction_t direction
,
2898 Object
**ret
, uint64_t *offset
) {
2906 /* First, seek by time */
2907 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, &b
);
2913 r
= generic_array_bisect_plus_one(f
,
2914 le64toh(o
->data
.entry_offset
),
2915 le64toh(o
->data
.entry_array_offset
),
2916 le64toh(o
->data
.n_entries
),
2918 test_object_monotonic
,
2924 /* And now, continue seeking until we find an entry that
2925 * exists in both bisection arrays */
2931 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2935 r
= generic_array_bisect_plus_one(f
,
2936 le64toh(d
->data
.entry_offset
),
2937 le64toh(d
->data
.entry_array_offset
),
2938 le64toh(d
->data
.n_entries
),
2946 r
= journal_file_move_to_object(f
, OBJECT_DATA
, b
, &o
);
2950 r
= generic_array_bisect_plus_one(f
,
2951 le64toh(o
->data
.entry_offset
),
2952 le64toh(o
->data
.entry_array_offset
),
2953 le64toh(o
->data
.n_entries
),
2975 int journal_file_move_to_entry_by_seqnum_for_data(
2977 uint64_t data_offset
,
2979 direction_t direction
,
2980 Object
**ret
, uint64_t *offset
) {
2987 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2991 return generic_array_bisect_plus_one(f
,
2992 le64toh(d
->data
.entry_offset
),
2993 le64toh(d
->data
.entry_array_offset
),
2994 le64toh(d
->data
.n_entries
),
3001 int journal_file_move_to_entry_by_realtime_for_data(
3003 uint64_t data_offset
,
3005 direction_t direction
,
3006 Object
**ret
, uint64_t *offset
) {
3013 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
3017 return generic_array_bisect_plus_one(f
,
3018 le64toh(d
->data
.entry_offset
),
3019 le64toh(d
->data
.entry_array_offset
),
3020 le64toh(d
->data
.n_entries
),
3022 test_object_realtime
,
3027 void journal_file_dump(JournalFile
*f
) {
3035 journal_file_print_header(f
);
3037 p
= le64toh(f
->header
->header_size
);
3039 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &o
);
3043 switch (o
->object
.type
) {
3046 printf("Type: OBJECT_UNUSED\n");
3050 printf("Type: OBJECT_DATA\n");
3054 printf("Type: OBJECT_FIELD\n");
3058 printf("Type: OBJECT_ENTRY seqnum=%"PRIu64
" monotonic=%"PRIu64
" realtime=%"PRIu64
"\n",
3059 le64toh(o
->entry
.seqnum
),
3060 le64toh(o
->entry
.monotonic
),
3061 le64toh(o
->entry
.realtime
));
3064 case OBJECT_FIELD_HASH_TABLE
:
3065 printf("Type: OBJECT_FIELD_HASH_TABLE\n");
3068 case OBJECT_DATA_HASH_TABLE
:
3069 printf("Type: OBJECT_DATA_HASH_TABLE\n");
3072 case OBJECT_ENTRY_ARRAY
:
3073 printf("Type: OBJECT_ENTRY_ARRAY\n");
3077 printf("Type: OBJECT_TAG seqnum=%"PRIu64
" epoch=%"PRIu64
"\n",
3078 le64toh(o
->tag
.seqnum
),
3079 le64toh(o
->tag
.epoch
));
3083 printf("Type: unknown (%i)\n", o
->object
.type
);
3087 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
)
3088 printf("Flags: %s\n",
3089 object_compressed_to_string(o
->object
.flags
& OBJECT_COMPRESSION_MASK
));
3091 if (p
== le64toh(f
->header
->tail_object_offset
))
3094 p
= p
+ ALIGN64(le64toh(o
->object
.size
));
3099 log_error("File corrupt");
3102 static const char* format_timestamp_safe(char *buf
, size_t l
, usec_t t
) {
3105 x
= format_timestamp(buf
, l
, t
);
3111 void journal_file_print_header(JournalFile
*f
) {
3112 char a
[33], b
[33], c
[33], d
[33];
3113 char x
[FORMAT_TIMESTAMP_MAX
], y
[FORMAT_TIMESTAMP_MAX
], z
[FORMAT_TIMESTAMP_MAX
];
3115 char bytes
[FORMAT_BYTES_MAX
];
3120 printf("File Path: %s\n"
3124 "Sequential Number ID: %s\n"
3126 "Compatible Flags:%s%s\n"
3127 "Incompatible Flags:%s%s%s\n"
3128 "Header size: %"PRIu64
"\n"
3129 "Arena size: %"PRIu64
"\n"
3130 "Data Hash Table Size: %"PRIu64
"\n"
3131 "Field Hash Table Size: %"PRIu64
"\n"
3132 "Rotate Suggested: %s\n"
3133 "Head Sequential Number: %"PRIu64
" (%"PRIx64
")\n"
3134 "Tail Sequential Number: %"PRIu64
" (%"PRIx64
")\n"
3135 "Head Realtime Timestamp: %s (%"PRIx64
")\n"
3136 "Tail Realtime Timestamp: %s (%"PRIx64
")\n"
3137 "Tail Monotonic Timestamp: %s (%"PRIx64
")\n"
3138 "Objects: %"PRIu64
"\n"
3139 "Entry Objects: %"PRIu64
"\n",
3141 sd_id128_to_string(f
->header
->file_id
, a
),
3142 sd_id128_to_string(f
->header
->machine_id
, b
),
3143 sd_id128_to_string(f
->header
->boot_id
, c
),
3144 sd_id128_to_string(f
->header
->seqnum_id
, d
),
3145 f
->header
->state
== STATE_OFFLINE
? "OFFLINE" :
3146 f
->header
->state
== STATE_ONLINE
? "ONLINE" :
3147 f
->header
->state
== STATE_ARCHIVED
? "ARCHIVED" : "UNKNOWN",
3148 JOURNAL_HEADER_SEALED(f
->header
) ? " SEALED" : "",
3149 (le32toh(f
->header
->compatible_flags
) & ~HEADER_COMPATIBLE_ANY
) ? " ???" : "",
3150 JOURNAL_HEADER_COMPRESSED_XZ(f
->header
) ? " COMPRESSED-XZ" : "",
3151 JOURNAL_HEADER_COMPRESSED_LZ4(f
->header
) ? " COMPRESSED-LZ4" : "",
3152 (le32toh(f
->header
->incompatible_flags
) & ~HEADER_INCOMPATIBLE_ANY
) ? " ???" : "",
3153 le64toh(f
->header
->header_size
),
3154 le64toh(f
->header
->arena_size
),
3155 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
3156 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
),
3157 yes_no(journal_file_rotate_suggested(f
, 0)),
3158 le64toh(f
->header
->head_entry_seqnum
), le64toh(f
->header
->head_entry_seqnum
),
3159 le64toh(f
->header
->tail_entry_seqnum
), le64toh(f
->header
->tail_entry_seqnum
),
3160 format_timestamp_safe(x
, sizeof(x
), le64toh(f
->header
->head_entry_realtime
)), le64toh(f
->header
->head_entry_realtime
),
3161 format_timestamp_safe(y
, sizeof(y
), le64toh(f
->header
->tail_entry_realtime
)), le64toh(f
->header
->tail_entry_realtime
),
3162 format_timespan(z
, sizeof(z
), le64toh(f
->header
->tail_entry_monotonic
), USEC_PER_MSEC
), le64toh(f
->header
->tail_entry_monotonic
),
3163 le64toh(f
->header
->n_objects
),
3164 le64toh(f
->header
->n_entries
));
3166 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
3167 printf("Data Objects: %"PRIu64
"\n"
3168 "Data Hash Table Fill: %.1f%%\n",
3169 le64toh(f
->header
->n_data
),
3170 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))));
3172 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
3173 printf("Field Objects: %"PRIu64
"\n"
3174 "Field Hash Table Fill: %.1f%%\n",
3175 le64toh(f
->header
->n_fields
),
3176 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))));
3178 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_tags
))
3179 printf("Tag Objects: %"PRIu64
"\n",
3180 le64toh(f
->header
->n_tags
));
3181 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
3182 printf("Entry Array Objects: %"PRIu64
"\n",
3183 le64toh(f
->header
->n_entry_arrays
));
3185 if (fstat(f
->fd
, &st
) >= 0)
3186 printf("Disk usage: %s\n", format_bytes(bytes
, sizeof(bytes
), (uint64_t) st
.st_blocks
* 512ULL));
3189 static int journal_file_warn_btrfs(JournalFile
*f
) {
3195 /* Before we write anything, check if the COW logic is turned
3196 * off on btrfs. Given our write pattern that is quite
3197 * unfriendly to COW file systems this should greatly improve
3198 * performance on COW file systems, such as btrfs, at the
3199 * expense of data integrity features (which shouldn't be too
3200 * bad, given that we do our own checksumming). */
3202 r
= btrfs_is_filesystem(f
->fd
);
3204 return log_warning_errno(r
, "Failed to determine if journal is on btrfs: %m");
3208 r
= read_attr_fd(f
->fd
, &attrs
);
3210 return log_warning_errno(r
, "Failed to read file attributes: %m");
3212 if (attrs
& FS_NOCOW_FL
) {
3213 log_debug("Detected btrfs file system with copy-on-write disabled, all is good.");
3217 log_notice("Creating journal file %s on a btrfs file system, and copy-on-write is enabled. "
3218 "This is likely to slow down journal access substantially, please consider turning "
3219 "off the copy-on-write file attribute on the journal directory, using chattr +C.", f
->path
);
3224 int journal_file_open(
3231 JournalMetrics
*metrics
,
3232 MMapCache
*mmap_cache
,
3233 Set
*deferred_closes
,
3234 JournalFile
*template,
3235 JournalFile
**ret
) {
3237 bool newly_created
= false;
3243 assert(fd
>= 0 || fname
);
3245 if ((flags
& O_ACCMODE
) != O_RDONLY
&&
3246 (flags
& O_ACCMODE
) != O_RDWR
)
3250 if (!endswith(fname
, ".journal") &&
3251 !endswith(fname
, ".journal~"))
3255 f
= new0(JournalFile
, 1);
3263 f
->prot
= prot_from_flags(flags
);
3264 f
->writable
= (flags
& O_ACCMODE
) != O_RDONLY
;
3265 #if defined(HAVE_LZ4)
3266 f
->compress_lz4
= compress
;
3267 #elif defined(HAVE_XZ)
3268 f
->compress_xz
= compress
;
3275 f
->mmap
= mmap_cache_ref(mmap_cache
);
3277 f
->mmap
= mmap_cache_new();
3285 f
->path
= strdup(fname
);
3291 /* If we don't know the path, fill in something explanatory and vaguely useful */
3292 if (asprintf(&f
->path
, "/proc/self/%i", fd
) < 0) {
3298 f
->chain_cache
= ordered_hashmap_new(&uint64_hash_ops
);
3299 if (!f
->chain_cache
) {
3305 f
->fd
= open(f
->path
, f
->flags
|O_CLOEXEC
, f
->mode
);
3311 /* fds we opened here by us should also be closed by us. */
3315 f
->cache_fd
= mmap_cache_add_fd(f
->mmap
, f
->fd
);
3321 r
= journal_file_fstat(f
);
3325 if (f
->last_stat
.st_size
== 0 && f
->writable
) {
3327 (void) journal_file_warn_btrfs(f
);
3329 /* Let's attach the creation time to the journal file,
3330 * so that the vacuuming code knows the age of this
3331 * file even if the file might end up corrupted one
3332 * day... Ideally we'd just use the creation time many
3333 * file systems maintain for each file, but there is
3334 * currently no usable API to query this, hence let's
3335 * emulate this via extended attributes. If extended
3336 * attributes are not supported we'll just skip this,
3337 * and rely solely on mtime/atime/ctime of the file. */
3339 fd_setcrtime(f
->fd
, 0);
3342 /* Try to load the FSPRG state, and if we can't, then
3343 * just don't do sealing */
3345 r
= journal_file_fss_load(f
);
3351 r
= journal_file_init_header(f
, template);
3355 r
= journal_file_fstat(f
);
3359 newly_created
= true;
3362 if (f
->last_stat
.st_size
< (off_t
) HEADER_SIZE_MIN
) {
3367 r
= mmap_cache_get(f
->mmap
, f
->cache_fd
, f
->prot
, CONTEXT_HEADER
, true, 0, PAGE_ALIGN(sizeof(Header
)), &f
->last_stat
, &h
, NULL
);
3373 if (!newly_created
) {
3374 if (deferred_closes
)
3375 journal_file_close_set(deferred_closes
);
3377 r
= journal_file_verify_header(f
);
3383 if (!newly_created
&& f
->writable
) {
3384 r
= journal_file_fss_load(f
);
3392 journal_default_metrics(metrics
, f
->fd
);
3393 f
->metrics
= *metrics
;
3394 } else if (template)
3395 f
->metrics
= template->metrics
;
3397 r
= journal_file_refresh_header(f
);
3403 r
= journal_file_hmac_setup(f
);
3408 if (newly_created
) {
3409 r
= journal_file_setup_field_hash_table(f
);
3413 r
= journal_file_setup_data_hash_table(f
);
3418 r
= journal_file_append_first_tag(f
);
3424 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
)) {
3429 if (template && template->post_change_timer
) {
3430 r
= journal_file_enable_post_change_timer(
3432 sd_event_source_get_event(template->post_change_timer
),
3433 template->post_change_timer_period
);
3439 /* The file is opened now successfully, thus we take possession of any passed in fd. */
3446 if (f
->cache_fd
&& mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
3449 (void) journal_file_close(f
);
3454 int journal_file_rotate(JournalFile
**f
, bool compress
, bool seal
, Set
*deferred_closes
) {
3455 _cleanup_free_
char *p
= NULL
;
3457 JournalFile
*old_file
, *new_file
= NULL
;
3465 if (!old_file
->writable
)
3468 /* Is this a journal file that was passed to us as fd? If so, we synthesized a path name for it, and we refuse
3469 * rotation, since we don't know the actual path, and couldn't rename the file hence. */
3470 if (path_startswith(old_file
->path
, "/proc/self/fd"))
3473 if (!endswith(old_file
->path
, ".journal"))
3476 l
= strlen(old_file
->path
);
3477 r
= asprintf(&p
, "%.*s@" SD_ID128_FORMAT_STR
"-%016"PRIx64
"-%016"PRIx64
".journal",
3478 (int) l
- 8, old_file
->path
,
3479 SD_ID128_FORMAT_VAL(old_file
->header
->seqnum_id
),
3480 le64toh((*f
)->header
->head_entry_seqnum
),
3481 le64toh((*f
)->header
->head_entry_realtime
));
3485 /* Try to rename the file to the archived version. If the file
3486 * already was deleted, we'll get ENOENT, let's ignore that
3488 r
= rename(old_file
->path
, p
);
3489 if (r
< 0 && errno
!= ENOENT
)
3492 /* Sync the rename to disk */
3493 (void) fsync_directory_of_file(old_file
->fd
);
3495 /* Set as archive so offlining commits w/state=STATE_ARCHIVED.
3496 * Previously we would set old_file->header->state to STATE_ARCHIVED directly here,
3497 * but journal_file_set_offline() short-circuits when state != STATE_ONLINE, which
3498 * would result in the rotated journal never getting fsync() called before closing.
3499 * Now we simply queue the archive state by setting an archive bit, leaving the state
3500 * as STATE_ONLINE so proper offlining occurs. */
3501 old_file
->archive
= true;
3503 /* Currently, btrfs is not very good with out write patterns
3504 * and fragments heavily. Let's defrag our journal files when
3505 * we archive them */
3506 old_file
->defrag_on_close
= true;
3508 r
= journal_file_open(-1, old_file
->path
, old_file
->flags
, old_file
->mode
, compress
, seal
, NULL
, old_file
->mmap
, deferred_closes
, old_file
, &new_file
);
3510 if (deferred_closes
&&
3511 set_put(deferred_closes
, old_file
) >= 0)
3512 (void) journal_file_set_offline(old_file
, false);
3514 (void) journal_file_close(old_file
);
3520 int journal_file_open_reliably(
3526 JournalMetrics
*metrics
,
3527 MMapCache
*mmap_cache
,
3528 Set
*deferred_closes
,
3529 JournalFile
*template,
3530 JournalFile
**ret
) {
3534 _cleanup_free_
char *p
= NULL
;
3536 r
= journal_file_open(-1, fname
, flags
, mode
, compress
, seal
, metrics
, mmap_cache
, deferred_closes
, template, ret
);
3538 -EBADMSG
, /* Corrupted */
3539 -ENODATA
, /* Truncated */
3540 -EHOSTDOWN
, /* Other machine */
3541 -EPROTONOSUPPORT
, /* Incompatible feature */
3542 -EBUSY
, /* Unclean shutdown */
3543 -ESHUTDOWN
, /* Already archived */
3544 -EIO
, /* IO error, including SIGBUS on mmap */
3545 -EIDRM
, /* File has been deleted */
3546 -ETXTBSY
)) /* File is from the future */
3549 if ((flags
& O_ACCMODE
) == O_RDONLY
)
3552 if (!(flags
& O_CREAT
))
3555 if (!endswith(fname
, ".journal"))
3558 /* The file is corrupted. Rotate it away and try it again (but only once) */
3561 if (asprintf(&p
, "%.*s@%016"PRIx64
"-%016"PRIx64
".journal~",
3563 now(CLOCK_REALTIME
),
3567 if (rename(fname
, p
) < 0)
3570 /* btrfs doesn't cope well with our write pattern and
3571 * fragments heavily. Let's defrag all files we rotate */
3573 (void) chattr_path(p
, 0, FS_NOCOW_FL
);
3574 (void) btrfs_defrag(p
);
3576 log_warning_errno(r
, "File %s corrupted or uncleanly shut down, renaming and replacing.", fname
);
3578 return journal_file_open(-1, fname
, flags
, mode
, compress
, seal
, metrics
, mmap_cache
, deferred_closes
, template, ret
);
3581 int journal_file_copy_entry(JournalFile
*from
, JournalFile
*to
, Object
*o
, uint64_t p
, uint64_t *seqnum
, Object
**ret
, uint64_t *offset
) {
3583 uint64_t q
, xor_hash
= 0;
3596 ts
.monotonic
= le64toh(o
->entry
.monotonic
);
3597 ts
.realtime
= le64toh(o
->entry
.realtime
);
3599 n
= journal_file_entry_n_items(o
);
3600 /* alloca() can't take 0, hence let's allocate at least one */
3601 items
= alloca(sizeof(EntryItem
) * MAX(1u, n
));
3603 for (i
= 0; i
< n
; i
++) {
3610 q
= le64toh(o
->entry
.items
[i
].object_offset
);
3611 le_hash
= o
->entry
.items
[i
].hash
;
3613 r
= journal_file_move_to_object(from
, OBJECT_DATA
, q
, &o
);
3617 if (le_hash
!= o
->data
.hash
)
3620 l
= le64toh(o
->object
.size
) - offsetof(Object
, data
.payload
);
3623 /* We hit the limit on 32bit machines */
3624 if ((uint64_t) t
!= l
)
3627 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
) {
3628 #if defined(HAVE_XZ) || defined(HAVE_LZ4)
3631 r
= decompress_blob(o
->object
.flags
& OBJECT_COMPRESSION_MASK
,
3632 o
->data
.payload
, l
, &from
->compress_buffer
, &from
->compress_buffer_size
, &rsize
, 0);
3636 data
= from
->compress_buffer
;
3639 return -EPROTONOSUPPORT
;
3642 data
= o
->data
.payload
;
3644 r
= journal_file_append_data(to
, data
, l
, &u
, &h
);
3648 xor_hash
^= le64toh(u
->data
.hash
);
3649 items
[i
].object_offset
= htole64(h
);
3650 items
[i
].hash
= u
->data
.hash
;
3652 r
= journal_file_move_to_object(from
, OBJECT_ENTRY
, p
, &o
);
3657 r
= journal_file_append_entry_internal(to
, &ts
, xor_hash
, items
, n
, seqnum
, ret
, offset
);
3659 if (mmap_cache_got_sigbus(to
->mmap
, to
->cache_fd
))
3665 void journal_reset_metrics(JournalMetrics
*m
) {
3668 /* Set everything to "pick automatic values". */
3670 *m
= (JournalMetrics
) {
3671 .min_use
= (uint64_t) -1,
3672 .max_use
= (uint64_t) -1,
3673 .min_size
= (uint64_t) -1,
3674 .max_size
= (uint64_t) -1,
3675 .keep_free
= (uint64_t) -1,
3676 .n_max_files
= (uint64_t) -1,
3680 void journal_default_metrics(JournalMetrics
*m
, int fd
) {
3681 char a
[FORMAT_BYTES_MAX
], b
[FORMAT_BYTES_MAX
], c
[FORMAT_BYTES_MAX
], d
[FORMAT_BYTES_MAX
], e
[FORMAT_BYTES_MAX
];
3688 if (fstatvfs(fd
, &ss
) >= 0)
3689 fs_size
= ss
.f_frsize
* ss
.f_blocks
;
3691 log_debug_errno(errno
, "Failed to detremine disk size: %m");
3695 if (m
->max_use
== (uint64_t) -1) {
3698 m
->max_use
= PAGE_ALIGN(fs_size
/ 10); /* 10% of file system size */
3700 if (m
->max_use
> DEFAULT_MAX_USE_UPPER
)
3701 m
->max_use
= DEFAULT_MAX_USE_UPPER
;
3703 if (m
->max_use
< DEFAULT_MAX_USE_LOWER
)
3704 m
->max_use
= DEFAULT_MAX_USE_LOWER
;
3706 m
->max_use
= DEFAULT_MAX_USE_LOWER
;
3708 m
->max_use
= PAGE_ALIGN(m
->max_use
);
3710 if (m
->max_use
!= 0 && m
->max_use
< JOURNAL_FILE_SIZE_MIN
*2)
3711 m
->max_use
= JOURNAL_FILE_SIZE_MIN
*2;
3714 if (m
->min_use
== (uint64_t) -1)
3715 m
->min_use
= DEFAULT_MIN_USE
;
3717 if (m
->min_use
> m
->max_use
)
3718 m
->min_use
= m
->max_use
;
3720 if (m
->max_size
== (uint64_t) -1) {
3721 m
->max_size
= PAGE_ALIGN(m
->max_use
/ 8); /* 8 chunks */
3723 if (m
->max_size
> DEFAULT_MAX_SIZE_UPPER
)
3724 m
->max_size
= DEFAULT_MAX_SIZE_UPPER
;
3726 m
->max_size
= PAGE_ALIGN(m
->max_size
);
3728 if (m
->max_size
!= 0) {
3729 if (m
->max_size
< JOURNAL_FILE_SIZE_MIN
)
3730 m
->max_size
= JOURNAL_FILE_SIZE_MIN
;
3732 if (m
->max_use
!= 0 && m
->max_size
*2 > m
->max_use
)
3733 m
->max_use
= m
->max_size
*2;
3736 if (m
->min_size
== (uint64_t) -1)
3737 m
->min_size
= JOURNAL_FILE_SIZE_MIN
;
3739 m
->min_size
= PAGE_ALIGN(m
->min_size
);
3741 if (m
->min_size
< JOURNAL_FILE_SIZE_MIN
)
3742 m
->min_size
= JOURNAL_FILE_SIZE_MIN
;
3744 if (m
->max_size
!= 0 && m
->min_size
> m
->max_size
)
3745 m
->max_size
= m
->min_size
;
3748 if (m
->keep_free
== (uint64_t) -1) {
3751 m
->keep_free
= PAGE_ALIGN(fs_size
* 3 / 20); /* 15% of file system size */
3753 if (m
->keep_free
> DEFAULT_KEEP_FREE_UPPER
)
3754 m
->keep_free
= DEFAULT_KEEP_FREE_UPPER
;
3757 m
->keep_free
= DEFAULT_KEEP_FREE
;
3760 if (m
->n_max_files
== (uint64_t) -1)
3761 m
->n_max_files
= DEFAULT_N_MAX_FILES
;
3763 log_debug("Fixed min_use=%s max_use=%s max_size=%s min_size=%s keep_free=%s n_max_files=%" PRIu64
,
3764 format_bytes(a
, sizeof(a
), m
->min_use
),
3765 format_bytes(b
, sizeof(b
), m
->max_use
),
3766 format_bytes(c
, sizeof(c
), m
->max_size
),
3767 format_bytes(d
, sizeof(d
), m
->min_size
),
3768 format_bytes(e
, sizeof(e
), m
->keep_free
),
3772 int journal_file_get_cutoff_realtime_usec(JournalFile
*f
, usec_t
*from
, usec_t
*to
) {
3778 if (f
->header
->head_entry_realtime
== 0)
3781 *from
= le64toh(f
->header
->head_entry_realtime
);
3785 if (f
->header
->tail_entry_realtime
== 0)
3788 *to
= le64toh(f
->header
->tail_entry_realtime
);
3794 int journal_file_get_cutoff_monotonic_usec(JournalFile
*f
, sd_id128_t boot_id
, usec_t
*from
, usec_t
*to
) {
3802 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, &p
);
3806 if (le64toh(o
->data
.n_entries
) <= 0)
3810 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, le64toh(o
->data
.entry_offset
), &o
);
3814 *from
= le64toh(o
->entry
.monotonic
);
3818 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
3822 r
= generic_array_get_plus_one(f
,
3823 le64toh(o
->data
.entry_offset
),
3824 le64toh(o
->data
.entry_array_offset
),
3825 le64toh(o
->data
.n_entries
)-1,
3830 *to
= le64toh(o
->entry
.monotonic
);
3836 bool journal_file_rotate_suggested(JournalFile
*f
, usec_t max_file_usec
) {
3840 /* If we gained new header fields we gained new features,
3841 * hence suggest a rotation */
3842 if (le64toh(f
->header
->header_size
) < sizeof(Header
)) {
3843 log_debug("%s uses an outdated header, suggesting rotation.", f
->path
);
3847 /* Let's check if the hash tables grew over a certain fill
3848 * level (75%, borrowing this value from Java's hash table
3849 * implementation), and if so suggest a rotation. To calculate
3850 * the fill level we need the n_data field, which only exists
3851 * in newer versions. */
3853 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
3854 if (le64toh(f
->header
->n_data
) * 4ULL > (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
3855 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.",
3857 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))),
3858 le64toh(f
->header
->n_data
),
3859 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
3860 (unsigned long long) f
->last_stat
.st_size
,
3861 f
->last_stat
.st_size
/ le64toh(f
->header
->n_data
));
3865 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
3866 if (le64toh(f
->header
->n_fields
) * 4ULL > (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
3867 log_debug("Field hash table of %s has a fill level at %.1f (%"PRIu64
" of %"PRIu64
" items), suggesting rotation.",
3869 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))),
3870 le64toh(f
->header
->n_fields
),
3871 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
));
3875 /* Are the data objects properly indexed by field objects? */
3876 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
) &&
3877 JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
) &&
3878 le64toh(f
->header
->n_data
) > 0 &&
3879 le64toh(f
->header
->n_fields
) == 0)
3882 if (max_file_usec
> 0) {
3885 h
= le64toh(f
->header
->head_entry_realtime
);
3886 t
= now(CLOCK_REALTIME
);
3888 if (h
> 0 && t
> h
+ max_file_usec
)