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
->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
->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
->fd
>= 0)
360 mmap_cache_close_fd(f
->mmap
, f
->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
) {
552 if (memcmp(f
->header
->signature
, HEADER_SIGNATURE
, 8))
555 /* In both read and write mode we refuse to open files with incompatible
556 * flags we don't know. */
557 if (warn_wrong_flags(f
, false))
558 return -EPROTONOSUPPORT
;
560 /* When open for writing we refuse to open files with compatible flags, too. */
561 if (f
->writable
&& warn_wrong_flags(f
, true))
562 return -EPROTONOSUPPORT
;
564 if (f
->header
->state
>= _STATE_MAX
)
567 /* The first addition was n_data, so check that we are at least this large */
568 if (le64toh(f
->header
->header_size
) < HEADER_SIZE_MIN
)
571 if (JOURNAL_HEADER_SEALED(f
->header
) && !JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
574 if ((le64toh(f
->header
->header_size
) + le64toh(f
->header
->arena_size
)) > (uint64_t) f
->last_stat
.st_size
)
577 if (le64toh(f
->header
->tail_object_offset
) > (le64toh(f
->header
->header_size
) + le64toh(f
->header
->arena_size
)))
580 if (!VALID64(le64toh(f
->header
->data_hash_table_offset
)) ||
581 !VALID64(le64toh(f
->header
->field_hash_table_offset
)) ||
582 !VALID64(le64toh(f
->header
->tail_object_offset
)) ||
583 !VALID64(le64toh(f
->header
->entry_array_offset
)))
587 sd_id128_t machine_id
;
591 r
= sd_id128_get_machine(&machine_id
);
595 if (!sd_id128_equal(machine_id
, f
->header
->machine_id
))
598 state
= f
->header
->state
;
600 if (state
== STATE_ARCHIVED
)
601 return -ESHUTDOWN
; /* Already archived */
602 else if (state
== STATE_ONLINE
) {
603 log_debug("Journal file %s is already online. Assuming unclean closing.", f
->path
);
605 } else if (state
!= STATE_OFFLINE
) {
606 log_debug("Journal file %s has unknown state %i.", f
->path
, state
);
610 /* Don't permit appending to files from the future. Because otherwise the realtime timestamps wouldn't
611 * be strictly ordered in the entries in the file anymore, and we can't have that since it breaks
613 if (le64toh(f
->header
->tail_entry_realtime
) > now(CLOCK_REALTIME
)) {
614 log_debug("Journal file %s is from the future, refusing to append new data to it that'd be older.", f
->path
);
619 f
->compress_xz
= JOURNAL_HEADER_COMPRESSED_XZ(f
->header
);
620 f
->compress_lz4
= JOURNAL_HEADER_COMPRESSED_LZ4(f
->header
);
622 f
->seal
= JOURNAL_HEADER_SEALED(f
->header
);
627 static int journal_file_fstat(JournalFile
*f
) {
631 if (fstat(f
->fd
, &f
->last_stat
) < 0)
634 f
->last_stat_usec
= now(CLOCK_MONOTONIC
);
636 /* Refuse appending to files that are already deleted */
637 if (f
->last_stat
.st_nlink
<= 0)
643 static int journal_file_allocate(JournalFile
*f
, uint64_t offset
, uint64_t size
) {
644 uint64_t old_size
, new_size
;
650 /* We assume that this file is not sparse, and we know that
651 * for sure, since we always call posix_fallocate()
654 if (mmap_cache_got_sigbus(f
->mmap
, f
->fd
))
658 le64toh(f
->header
->header_size
) +
659 le64toh(f
->header
->arena_size
);
661 new_size
= PAGE_ALIGN(offset
+ size
);
662 if (new_size
< le64toh(f
->header
->header_size
))
663 new_size
= le64toh(f
->header
->header_size
);
665 if (new_size
<= old_size
) {
667 /* We already pre-allocated enough space, but before
668 * we write to it, let's check with fstat() if the
669 * file got deleted, in order make sure we don't throw
670 * away the data immediately. Don't check fstat() for
671 * all writes though, but only once ever 10s. */
673 if (f
->last_stat_usec
+ LAST_STAT_REFRESH_USEC
> now(CLOCK_MONOTONIC
))
676 return journal_file_fstat(f
);
679 /* Allocate more space. */
681 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
684 if (new_size
> f
->metrics
.min_size
&& f
->metrics
.keep_free
> 0) {
687 if (fstatvfs(f
->fd
, &svfs
) >= 0) {
690 available
= LESS_BY((uint64_t) svfs
.f_bfree
* (uint64_t) svfs
.f_bsize
, f
->metrics
.keep_free
);
692 if (new_size
- old_size
> available
)
697 /* Increase by larger blocks at once */
698 new_size
= ((new_size
+FILE_SIZE_INCREASE
-1) / FILE_SIZE_INCREASE
) * FILE_SIZE_INCREASE
;
699 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
700 new_size
= f
->metrics
.max_size
;
702 /* Note that the glibc fallocate() fallback is very
703 inefficient, hence we try to minimize the allocation area
705 r
= posix_fallocate(f
->fd
, old_size
, new_size
- old_size
);
709 f
->header
->arena_size
= htole64(new_size
- le64toh(f
->header
->header_size
));
711 return journal_file_fstat(f
);
714 static unsigned type_to_context(ObjectType type
) {
715 /* One context for each type, plus one catch-all for the rest */
716 assert_cc(_OBJECT_TYPE_MAX
<= MMAP_CACHE_MAX_CONTEXTS
);
717 assert_cc(CONTEXT_HEADER
< MMAP_CACHE_MAX_CONTEXTS
);
718 return type
> OBJECT_UNUSED
&& type
< _OBJECT_TYPE_MAX
? type
: 0;
721 static int journal_file_move_to(JournalFile
*f
, ObjectType type
, bool keep_always
, uint64_t offset
, uint64_t size
, void **ret
) {
730 /* Avoid SIGBUS on invalid accesses */
731 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
) {
732 /* Hmm, out of range? Let's refresh the fstat() data
733 * first, before we trust that check. */
735 r
= journal_file_fstat(f
);
739 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
)
740 return -EADDRNOTAVAIL
;
743 return mmap_cache_get(f
->mmap
, f
->fd
, f
->prot
, type_to_context(type
), keep_always
, offset
, size
, &f
->last_stat
, ret
);
746 static uint64_t minimum_header_size(Object
*o
) {
748 static const uint64_t table
[] = {
749 [OBJECT_DATA
] = sizeof(DataObject
),
750 [OBJECT_FIELD
] = sizeof(FieldObject
),
751 [OBJECT_ENTRY
] = sizeof(EntryObject
),
752 [OBJECT_DATA_HASH_TABLE
] = sizeof(HashTableObject
),
753 [OBJECT_FIELD_HASH_TABLE
] = sizeof(HashTableObject
),
754 [OBJECT_ENTRY_ARRAY
] = sizeof(EntryArrayObject
),
755 [OBJECT_TAG
] = sizeof(TagObject
),
758 if (o
->object
.type
>= ELEMENTSOF(table
) || table
[o
->object
.type
] <= 0)
759 return sizeof(ObjectHeader
);
761 return table
[o
->object
.type
];
764 int journal_file_move_to_object(JournalFile
*f
, ObjectType type
, uint64_t offset
, Object
**ret
) {
773 /* Objects may only be located at multiple of 64 bit */
774 if (!VALID64(offset
)) {
775 log_debug("Attempt to move to object at non-64bit boundary: %" PRIu64
, offset
);
779 /* Object may not be located in the file header */
780 if (offset
< le64toh(f
->header
->header_size
)) {
781 log_debug("Attempt to move to object located in file header: %" PRIu64
, offset
);
785 r
= journal_file_move_to(f
, type
, false, offset
, sizeof(ObjectHeader
), &t
);
790 s
= le64toh(o
->object
.size
);
793 log_debug("Attempt to move to uninitialized object: %" PRIu64
, offset
);
796 if (s
< sizeof(ObjectHeader
)) {
797 log_debug("Attempt to move to overly short object: %" PRIu64
, offset
);
801 if (o
->object
.type
<= OBJECT_UNUSED
) {
802 log_debug("Attempt to move to object with invalid type: %" PRIu64
, offset
);
806 if (s
< minimum_header_size(o
)) {
807 log_debug("Attempt to move to truncated object: %" PRIu64
, offset
);
811 if (type
> OBJECT_UNUSED
&& o
->object
.type
!= type
) {
812 log_debug("Attempt to move to object of unexpected type: %" PRIu64
, offset
);
816 if (s
> sizeof(ObjectHeader
)) {
817 r
= journal_file_move_to(f
, type
, false, offset
, s
, &t
);
828 static uint64_t journal_file_entry_seqnum(JournalFile
*f
, uint64_t *seqnum
) {
834 r
= le64toh(f
->header
->tail_entry_seqnum
) + 1;
837 /* If an external seqnum counter was passed, we update
838 * both the local and the external one, and set it to
839 * the maximum of both */
847 f
->header
->tail_entry_seqnum
= htole64(r
);
849 if (f
->header
->head_entry_seqnum
== 0)
850 f
->header
->head_entry_seqnum
= htole64(r
);
855 int journal_file_append_object(JournalFile
*f
, ObjectType type
, uint64_t size
, Object
**ret
, uint64_t *offset
) {
863 assert(type
> OBJECT_UNUSED
&& type
< _OBJECT_TYPE_MAX
);
864 assert(size
>= sizeof(ObjectHeader
));
868 r
= journal_file_set_online(f
);
872 p
= le64toh(f
->header
->tail_object_offset
);
874 p
= le64toh(f
->header
->header_size
);
876 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &tail
);
880 p
+= ALIGN64(le64toh(tail
->object
.size
));
883 r
= journal_file_allocate(f
, p
, size
);
887 r
= journal_file_move_to(f
, type
, false, p
, size
, &t
);
894 o
->object
.type
= type
;
895 o
->object
.size
= htole64(size
);
897 f
->header
->tail_object_offset
= htole64(p
);
898 f
->header
->n_objects
= htole64(le64toh(f
->header
->n_objects
) + 1);
906 static int journal_file_setup_data_hash_table(JournalFile
*f
) {
914 /* We estimate that we need 1 hash table entry per 768 bytes
915 of journal file and we want to make sure we never get
916 beyond 75% fill level. Calculate the hash table size for
917 the maximum file size based on these metrics. */
919 s
= (f
->metrics
.max_size
* 4 / 768 / 3) * sizeof(HashItem
);
920 if (s
< DEFAULT_DATA_HASH_TABLE_SIZE
)
921 s
= DEFAULT_DATA_HASH_TABLE_SIZE
;
923 log_debug("Reserving %"PRIu64
" entries in hash table.", s
/ sizeof(HashItem
));
925 r
= journal_file_append_object(f
,
926 OBJECT_DATA_HASH_TABLE
,
927 offsetof(Object
, hash_table
.items
) + s
,
932 memzero(o
->hash_table
.items
, s
);
934 f
->header
->data_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
935 f
->header
->data_hash_table_size
= htole64(s
);
940 static int journal_file_setup_field_hash_table(JournalFile
*f
) {
948 /* We use a fixed size hash table for the fields as this
949 * number should grow very slowly only */
951 s
= DEFAULT_FIELD_HASH_TABLE_SIZE
;
952 r
= journal_file_append_object(f
,
953 OBJECT_FIELD_HASH_TABLE
,
954 offsetof(Object
, hash_table
.items
) + s
,
959 memzero(o
->hash_table
.items
, s
);
961 f
->header
->field_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
962 f
->header
->field_hash_table_size
= htole64(s
);
967 int journal_file_map_data_hash_table(JournalFile
*f
) {
975 if (f
->data_hash_table
)
978 p
= le64toh(f
->header
->data_hash_table_offset
);
979 s
= le64toh(f
->header
->data_hash_table_size
);
981 r
= journal_file_move_to(f
,
982 OBJECT_DATA_HASH_TABLE
,
989 f
->data_hash_table
= t
;
993 int journal_file_map_field_hash_table(JournalFile
*f
) {
1001 if (f
->field_hash_table
)
1004 p
= le64toh(f
->header
->field_hash_table_offset
);
1005 s
= le64toh(f
->header
->field_hash_table_size
);
1007 r
= journal_file_move_to(f
,
1008 OBJECT_FIELD_HASH_TABLE
,
1015 f
->field_hash_table
= t
;
1019 static int journal_file_link_field(
1030 assert(f
->field_hash_table
);
1034 if (o
->object
.type
!= OBJECT_FIELD
)
1037 m
= le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
);
1041 /* This might alter the window we are looking at */
1042 o
->field
.next_hash_offset
= o
->field
.head_data_offset
= 0;
1045 p
= le64toh(f
->field_hash_table
[h
].tail_hash_offset
);
1047 f
->field_hash_table
[h
].head_hash_offset
= htole64(offset
);
1049 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1053 o
->field
.next_hash_offset
= htole64(offset
);
1056 f
->field_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1058 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
1059 f
->header
->n_fields
= htole64(le64toh(f
->header
->n_fields
) + 1);
1064 static int journal_file_link_data(
1075 assert(f
->data_hash_table
);
1079 if (o
->object
.type
!= OBJECT_DATA
)
1082 m
= le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
);
1086 /* This might alter the window we are looking at */
1087 o
->data
.next_hash_offset
= o
->data
.next_field_offset
= 0;
1088 o
->data
.entry_offset
= o
->data
.entry_array_offset
= 0;
1089 o
->data
.n_entries
= 0;
1092 p
= le64toh(f
->data_hash_table
[h
].tail_hash_offset
);
1094 /* Only entry in the hash table is easy */
1095 f
->data_hash_table
[h
].head_hash_offset
= htole64(offset
);
1097 /* Move back to the previous data object, to patch in
1100 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1104 o
->data
.next_hash_offset
= htole64(offset
);
1107 f
->data_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1109 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
1110 f
->header
->n_data
= htole64(le64toh(f
->header
->n_data
) + 1);
1115 int journal_file_find_field_object_with_hash(
1117 const void *field
, uint64_t size
, uint64_t hash
,
1118 Object
**ret
, uint64_t *offset
) {
1120 uint64_t p
, osize
, h
, m
;
1125 assert(field
&& size
> 0);
1127 /* If the field hash table is empty, we can't find anything */
1128 if (le64toh(f
->header
->field_hash_table_size
) <= 0)
1131 /* Map the field hash table, if it isn't mapped yet. */
1132 r
= journal_file_map_field_hash_table(f
);
1136 osize
= offsetof(Object
, field
.payload
) + size
;
1138 m
= le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
);
1143 p
= le64toh(f
->field_hash_table
[h
].head_hash_offset
);
1148 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1152 if (le64toh(o
->field
.hash
) == hash
&&
1153 le64toh(o
->object
.size
) == osize
&&
1154 memcmp(o
->field
.payload
, field
, size
) == 0) {
1164 p
= le64toh(o
->field
.next_hash_offset
);
1170 int journal_file_find_field_object(
1172 const void *field
, uint64_t size
,
1173 Object
**ret
, uint64_t *offset
) {
1178 assert(field
&& size
> 0);
1180 hash
= hash64(field
, size
);
1182 return journal_file_find_field_object_with_hash(f
,
1187 int journal_file_find_data_object_with_hash(
1189 const void *data
, uint64_t size
, uint64_t hash
,
1190 Object
**ret
, uint64_t *offset
) {
1192 uint64_t p
, osize
, h
, m
;
1197 assert(data
|| size
== 0);
1199 /* If there's no data hash table, then there's no entry. */
1200 if (le64toh(f
->header
->data_hash_table_size
) <= 0)
1203 /* Map the data hash table, if it isn't mapped yet. */
1204 r
= journal_file_map_data_hash_table(f
);
1208 osize
= offsetof(Object
, data
.payload
) + size
;
1210 m
= le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
);
1215 p
= le64toh(f
->data_hash_table
[h
].head_hash_offset
);
1220 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1224 if (le64toh(o
->data
.hash
) != hash
)
1227 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
) {
1228 #if defined(HAVE_XZ) || defined(HAVE_LZ4)
1232 l
= le64toh(o
->object
.size
);
1233 if (l
<= offsetof(Object
, data
.payload
))
1236 l
-= offsetof(Object
, data
.payload
);
1238 r
= decompress_blob(o
->object
.flags
& OBJECT_COMPRESSION_MASK
,
1239 o
->data
.payload
, l
, &f
->compress_buffer
, &f
->compress_buffer_size
, &rsize
, 0);
1243 if (rsize
== size
&&
1244 memcmp(f
->compress_buffer
, data
, size
) == 0) {
1255 return -EPROTONOSUPPORT
;
1257 } else if (le64toh(o
->object
.size
) == osize
&&
1258 memcmp(o
->data
.payload
, data
, size
) == 0) {
1270 p
= le64toh(o
->data
.next_hash_offset
);
1276 int journal_file_find_data_object(
1278 const void *data
, uint64_t size
,
1279 Object
**ret
, uint64_t *offset
) {
1284 assert(data
|| size
== 0);
1286 hash
= hash64(data
, size
);
1288 return journal_file_find_data_object_with_hash(f
,
1293 static int journal_file_append_field(
1295 const void *field
, uint64_t size
,
1296 Object
**ret
, uint64_t *offset
) {
1304 assert(field
&& size
> 0);
1306 hash
= hash64(field
, size
);
1308 r
= journal_file_find_field_object_with_hash(f
, field
, size
, hash
, &o
, &p
);
1322 osize
= offsetof(Object
, field
.payload
) + size
;
1323 r
= journal_file_append_object(f
, OBJECT_FIELD
, osize
, &o
, &p
);
1327 o
->field
.hash
= htole64(hash
);
1328 memcpy(o
->field
.payload
, field
, size
);
1330 r
= journal_file_link_field(f
, o
, p
, hash
);
1334 /* The linking might have altered the window, so let's
1335 * refresh our pointer */
1336 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1341 r
= journal_file_hmac_put_object(f
, OBJECT_FIELD
, o
, p
);
1355 static int journal_file_append_data(
1357 const void *data
, uint64_t size
,
1358 Object
**ret
, uint64_t *offset
) {
1363 int r
, compression
= 0;
1367 assert(data
|| size
== 0);
1369 hash
= hash64(data
, size
);
1371 r
= journal_file_find_data_object_with_hash(f
, data
, size
, hash
, &o
, &p
);
1385 osize
= offsetof(Object
, data
.payload
) + size
;
1386 r
= journal_file_append_object(f
, OBJECT_DATA
, osize
, &o
, &p
);
1390 o
->data
.hash
= htole64(hash
);
1392 #if defined(HAVE_XZ) || defined(HAVE_LZ4)
1393 if (JOURNAL_FILE_COMPRESS(f
) && size
>= COMPRESSION_SIZE_THRESHOLD
) {
1396 compression
= compress_blob(data
, size
, o
->data
.payload
, size
- 1, &rsize
);
1398 if (compression
>= 0) {
1399 o
->object
.size
= htole64(offsetof(Object
, data
.payload
) + rsize
);
1400 o
->object
.flags
|= compression
;
1402 log_debug("Compressed data object %"PRIu64
" -> %zu using %s",
1403 size
, rsize
, object_compressed_to_string(compression
));
1405 /* Compression didn't work, we don't really care why, let's continue without compression */
1410 if (compression
== 0)
1411 memcpy_safe(o
->data
.payload
, data
, size
);
1413 r
= journal_file_link_data(f
, o
, p
, hash
);
1418 r
= journal_file_hmac_put_object(f
, OBJECT_DATA
, o
, p
);
1423 /* The linking might have altered the window, so let's
1424 * refresh our pointer */
1425 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1432 eq
= memchr(data
, '=', size
);
1433 if (eq
&& eq
> data
) {
1437 /* Create field object ... */
1438 r
= journal_file_append_field(f
, data
, (uint8_t*) eq
- (uint8_t*) data
, &fo
, &fp
);
1442 /* ... and link it in. */
1443 o
->data
.next_field_offset
= fo
->field
.head_data_offset
;
1444 fo
->field
.head_data_offset
= le64toh(p
);
1456 uint64_t journal_file_entry_n_items(Object
*o
) {
1459 if (o
->object
.type
!= OBJECT_ENTRY
)
1462 return (le64toh(o
->object
.size
) - offsetof(Object
, entry
.items
)) / sizeof(EntryItem
);
1465 uint64_t journal_file_entry_array_n_items(Object
*o
) {
1468 if (o
->object
.type
!= OBJECT_ENTRY_ARRAY
)
1471 return (le64toh(o
->object
.size
) - offsetof(Object
, entry_array
.items
)) / sizeof(uint64_t);
1474 uint64_t journal_file_hash_table_n_items(Object
*o
) {
1477 if (o
->object
.type
!= OBJECT_DATA_HASH_TABLE
&&
1478 o
->object
.type
!= OBJECT_FIELD_HASH_TABLE
)
1481 return (le64toh(o
->object
.size
) - offsetof(Object
, hash_table
.items
)) / sizeof(HashItem
);
1484 static int link_entry_into_array(JournalFile
*f
,
1489 uint64_t n
= 0, ap
= 0, q
, i
, a
, hidx
;
1498 a
= le64toh(*first
);
1499 i
= hidx
= le64toh(*idx
);
1502 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
1506 n
= journal_file_entry_array_n_items(o
);
1508 o
->entry_array
.items
[i
] = htole64(p
);
1509 *idx
= htole64(hidx
+ 1);
1515 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
1526 r
= journal_file_append_object(f
, OBJECT_ENTRY_ARRAY
,
1527 offsetof(Object
, entry_array
.items
) + n
* sizeof(uint64_t),
1533 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY_ARRAY
, o
, q
);
1538 o
->entry_array
.items
[i
] = htole64(p
);
1541 *first
= htole64(q
);
1543 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, ap
, &o
);
1547 o
->entry_array
.next_entry_array_offset
= htole64(q
);
1550 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
1551 f
->header
->n_entry_arrays
= htole64(le64toh(f
->header
->n_entry_arrays
) + 1);
1553 *idx
= htole64(hidx
+ 1);
1558 static int link_entry_into_array_plus_one(JournalFile
*f
,
1573 *extra
= htole64(p
);
1577 i
= htole64(le64toh(*idx
) - 1);
1578 r
= link_entry_into_array(f
, first
, &i
, p
);
1583 *idx
= htole64(le64toh(*idx
) + 1);
1587 static int journal_file_link_entry_item(JournalFile
*f
, Object
*o
, uint64_t offset
, uint64_t i
) {
1594 p
= le64toh(o
->entry
.items
[i
].object_offset
);
1598 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1602 return link_entry_into_array_plus_one(f
,
1603 &o
->data
.entry_offset
,
1604 &o
->data
.entry_array_offset
,
1609 static int journal_file_link_entry(JournalFile
*f
, Object
*o
, uint64_t offset
) {
1618 if (o
->object
.type
!= OBJECT_ENTRY
)
1621 __sync_synchronize();
1623 /* Link up the entry itself */
1624 r
= link_entry_into_array(f
,
1625 &f
->header
->entry_array_offset
,
1626 &f
->header
->n_entries
,
1631 /* log_debug("=> %s seqnr=%"PRIu64" n_entries=%"PRIu64, f->path, o->entry.seqnum, f->header->n_entries); */
1633 if (f
->header
->head_entry_realtime
== 0)
1634 f
->header
->head_entry_realtime
= o
->entry
.realtime
;
1636 f
->header
->tail_entry_realtime
= o
->entry
.realtime
;
1637 f
->header
->tail_entry_monotonic
= o
->entry
.monotonic
;
1639 f
->tail_entry_monotonic_valid
= true;
1641 /* Link up the items */
1642 n
= journal_file_entry_n_items(o
);
1643 for (i
= 0; i
< n
; i
++) {
1644 r
= journal_file_link_entry_item(f
, o
, offset
, i
);
1652 static int journal_file_append_entry_internal(
1654 const dual_timestamp
*ts
,
1656 const EntryItem items
[], unsigned n_items
,
1658 Object
**ret
, uint64_t *offset
) {
1666 assert(items
|| n_items
== 0);
1669 osize
= offsetof(Object
, entry
.items
) + (n_items
* sizeof(EntryItem
));
1671 r
= journal_file_append_object(f
, OBJECT_ENTRY
, osize
, &o
, &np
);
1675 o
->entry
.seqnum
= htole64(journal_file_entry_seqnum(f
, seqnum
));
1676 memcpy_safe(o
->entry
.items
, items
, n_items
* sizeof(EntryItem
));
1677 o
->entry
.realtime
= htole64(ts
->realtime
);
1678 o
->entry
.monotonic
= htole64(ts
->monotonic
);
1679 o
->entry
.xor_hash
= htole64(xor_hash
);
1680 o
->entry
.boot_id
= f
->header
->boot_id
;
1683 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY
, o
, np
);
1688 r
= journal_file_link_entry(f
, o
, np
);
1701 void journal_file_post_change(JournalFile
*f
) {
1704 /* inotify() does not receive IN_MODIFY events from file
1705 * accesses done via mmap(). After each access we hence
1706 * trigger IN_MODIFY by truncating the journal file to its
1707 * current size which triggers IN_MODIFY. */
1709 __sync_synchronize();
1711 if (ftruncate(f
->fd
, f
->last_stat
.st_size
) < 0)
1712 log_debug_errno(errno
, "Failed to truncate file to its own size: %m");
1715 static int post_change_thunk(sd_event_source
*timer
, uint64_t usec
, void *userdata
) {
1718 journal_file_post_change(userdata
);
1723 static void schedule_post_change(JournalFile
*f
) {
1724 sd_event_source
*timer
;
1729 assert(f
->post_change_timer
);
1731 timer
= f
->post_change_timer
;
1733 r
= sd_event_source_get_enabled(timer
, &enabled
);
1735 log_debug_errno(r
, "Failed to get ftruncate timer state: %m");
1739 if (enabled
== SD_EVENT_ONESHOT
)
1742 r
= sd_event_now(sd_event_source_get_event(timer
), CLOCK_MONOTONIC
, &now
);
1744 log_debug_errno(r
, "Failed to get clock's now for scheduling ftruncate: %m");
1748 r
= sd_event_source_set_time(timer
, now
+f
->post_change_timer_period
);
1750 log_debug_errno(r
, "Failed to set time for scheduling ftruncate: %m");
1754 r
= sd_event_source_set_enabled(timer
, SD_EVENT_ONESHOT
);
1756 log_debug_errno(r
, "Failed to enable scheduled ftruncate: %m");
1763 /* On failure, let's simply post the change immediately. */
1764 journal_file_post_change(f
);
1767 /* Enable coalesced change posting in a timer on the provided sd_event instance */
1768 int journal_file_enable_post_change_timer(JournalFile
*f
, sd_event
*e
, usec_t t
) {
1769 _cleanup_(sd_event_source_unrefp
) sd_event_source
*timer
= NULL
;
1773 assert_return(!f
->post_change_timer
, -EINVAL
);
1777 r
= sd_event_add_time(e
, &timer
, CLOCK_MONOTONIC
, 0, 0, post_change_thunk
, f
);
1781 r
= sd_event_source_set_enabled(timer
, SD_EVENT_OFF
);
1785 f
->post_change_timer
= timer
;
1787 f
->post_change_timer_period
= t
;
1792 static int entry_item_cmp(const void *_a
, const void *_b
) {
1793 const EntryItem
*a
= _a
, *b
= _b
;
1795 if (le64toh(a
->object_offset
) < le64toh(b
->object_offset
))
1797 if (le64toh(a
->object_offset
) > le64toh(b
->object_offset
))
1802 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
) {
1806 uint64_t xor_hash
= 0;
1807 struct dual_timestamp _ts
;
1811 assert(iovec
|| n_iovec
== 0);
1814 dual_timestamp_get(&_ts
);
1819 r
= journal_file_maybe_append_tag(f
, ts
->realtime
);
1824 /* alloca() can't take 0, hence let's allocate at least one */
1825 items
= alloca(sizeof(EntryItem
) * MAX(1u, n_iovec
));
1827 for (i
= 0; i
< n_iovec
; i
++) {
1831 r
= journal_file_append_data(f
, iovec
[i
].iov_base
, iovec
[i
].iov_len
, &o
, &p
);
1835 xor_hash
^= le64toh(o
->data
.hash
);
1836 items
[i
].object_offset
= htole64(p
);
1837 items
[i
].hash
= o
->data
.hash
;
1840 /* Order by the position on disk, in order to improve seek
1841 * times for rotating media. */
1842 qsort_safe(items
, n_iovec
, sizeof(EntryItem
), entry_item_cmp
);
1844 r
= journal_file_append_entry_internal(f
, ts
, xor_hash
, items
, n_iovec
, seqnum
, ret
, offset
);
1846 /* If the memory mapping triggered a SIGBUS then we return an
1847 * IO error and ignore the error code passed down to us, since
1848 * it is very likely just an effect of a nullified replacement
1851 if (mmap_cache_got_sigbus(f
->mmap
, f
->fd
))
1854 if (f
->post_change_timer
)
1855 schedule_post_change(f
);
1857 journal_file_post_change(f
);
1862 typedef struct ChainCacheItem
{
1863 uint64_t first
; /* the array at the beginning of the chain */
1864 uint64_t array
; /* the cached array */
1865 uint64_t begin
; /* the first item in the cached array */
1866 uint64_t total
; /* the total number of items in all arrays before this one in the chain */
1867 uint64_t last_index
; /* the last index we looked at, to optimize locality when bisecting */
1870 static void chain_cache_put(
1877 uint64_t last_index
) {
1880 /* If the chain item to cache for this chain is the
1881 * first one it's not worth caching anything */
1885 if (ordered_hashmap_size(h
) >= CHAIN_CACHE_MAX
) {
1886 ci
= ordered_hashmap_steal_first(h
);
1889 ci
= new(ChainCacheItem
, 1);
1896 if (ordered_hashmap_put(h
, &ci
->first
, ci
) < 0) {
1901 assert(ci
->first
== first
);
1906 ci
->last_index
= last_index
;
1909 static int generic_array_get(
1913 Object
**ret
, uint64_t *offset
) {
1916 uint64_t p
= 0, a
, t
= 0;
1924 /* Try the chain cache first */
1925 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
1926 if (ci
&& i
> ci
->total
) {
1935 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
1939 k
= journal_file_entry_array_n_items(o
);
1941 p
= le64toh(o
->entry_array
.items
[i
]);
1947 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
1953 /* Let's cache this item for the next invocation */
1954 chain_cache_put(f
->chain_cache
, ci
, first
, a
, le64toh(o
->entry_array
.items
[0]), t
, i
);
1956 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
1969 static int generic_array_get_plus_one(
1974 Object
**ret
, uint64_t *offset
) {
1983 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, &o
);
1996 return generic_array_get(f
, first
, i
-1, ret
, offset
);
2005 static int generic_array_bisect(
2010 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
2011 direction_t direction
,
2016 uint64_t a
, p
, t
= 0, i
= 0, last_p
= 0, last_index
= (uint64_t) -1;
2017 bool subtract_one
= false;
2018 Object
*o
, *array
= NULL
;
2023 assert(test_object
);
2025 /* Start with the first array in the chain */
2028 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
2029 if (ci
&& n
> ci
->total
) {
2030 /* Ah, we have iterated this bisection array chain
2031 * previously! Let's see if we can skip ahead in the
2032 * chain, as far as the last time. But we can't jump
2033 * backwards in the chain, so let's check that
2036 r
= test_object(f
, ci
->begin
, needle
);
2040 if (r
== TEST_LEFT
) {
2041 /* OK, what we are looking for is right of the
2042 * begin of this EntryArray, so let's jump
2043 * straight to previously cached array in the
2049 last_index
= ci
->last_index
;
2054 uint64_t left
, right
, k
, lp
;
2056 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &array
);
2060 k
= journal_file_entry_array_n_items(array
);
2066 lp
= p
= le64toh(array
->entry_array
.items
[i
]);
2070 r
= test_object(f
, p
, needle
);
2071 if (r
== -EBADMSG
) {
2072 log_debug_errno(r
, "Encountered invalid entry while bisecting, cutting algorithm short. (1)");
2079 if (r
== TEST_FOUND
)
2080 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2082 if (r
== TEST_RIGHT
) {
2086 if (last_index
!= (uint64_t) -1) {
2087 assert(last_index
<= right
);
2089 /* If we cached the last index we
2090 * looked at, let's try to not to jump
2091 * too wildly around and see if we can
2092 * limit the range to look at early to
2093 * the immediate neighbors of the last
2094 * index we looked at. */
2096 if (last_index
> 0) {
2097 uint64_t x
= last_index
- 1;
2099 p
= le64toh(array
->entry_array
.items
[x
]);
2103 r
= test_object(f
, p
, needle
);
2107 if (r
== TEST_FOUND
)
2108 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2110 if (r
== TEST_RIGHT
)
2116 if (last_index
< right
) {
2117 uint64_t y
= last_index
+ 1;
2119 p
= le64toh(array
->entry_array
.items
[y
]);
2123 r
= test_object(f
, p
, needle
);
2127 if (r
== TEST_FOUND
)
2128 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2130 if (r
== TEST_RIGHT
)
2138 if (left
== right
) {
2139 if (direction
== DIRECTION_UP
)
2140 subtract_one
= true;
2146 assert(left
< right
);
2147 i
= (left
+ right
) / 2;
2149 p
= le64toh(array
->entry_array
.items
[i
]);
2153 r
= test_object(f
, p
, needle
);
2154 if (r
== -EBADMSG
) {
2155 log_debug_errno(r
, "Encountered invalid entry while bisecting, cutting algorithm short. (2)");
2162 if (r
== TEST_FOUND
)
2163 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2165 if (r
== TEST_RIGHT
)
2173 if (direction
== DIRECTION_UP
) {
2175 subtract_one
= true;
2186 last_index
= (uint64_t) -1;
2187 a
= le64toh(array
->entry_array
.next_entry_array_offset
);
2193 if (subtract_one
&& t
== 0 && i
== 0)
2196 /* Let's cache this item for the next invocation */
2197 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
);
2199 if (subtract_one
&& i
== 0)
2201 else if (subtract_one
)
2202 p
= le64toh(array
->entry_array
.items
[i
-1]);
2204 p
= le64toh(array
->entry_array
.items
[i
]);
2206 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2217 *idx
= t
+ i
+ (subtract_one
? -1 : 0);
2222 static int generic_array_bisect_plus_one(
2228 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
2229 direction_t direction
,
2235 bool step_back
= false;
2239 assert(test_object
);
2244 /* This bisects the array in object 'first', but first checks
2246 r
= test_object(f
, extra
, needle
);
2250 if (r
== TEST_FOUND
)
2251 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2253 /* if we are looking with DIRECTION_UP then we need to first
2254 see if in the actual array there is a matching entry, and
2255 return the last one of that. But if there isn't any we need
2256 to return this one. Hence remember this, and return it
2259 step_back
= direction
== DIRECTION_UP
;
2261 if (r
== TEST_RIGHT
) {
2262 if (direction
== DIRECTION_DOWN
)
2268 r
= generic_array_bisect(f
, first
, n
-1, needle
, test_object
, direction
, ret
, offset
, idx
);
2270 if (r
== 0 && step_back
)
2279 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, &o
);
2295 _pure_
static int test_object_offset(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2301 else if (p
< needle
)
2307 static int test_object_seqnum(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2314 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2318 if (le64toh(o
->entry
.seqnum
) == needle
)
2320 else if (le64toh(o
->entry
.seqnum
) < needle
)
2326 int journal_file_move_to_entry_by_seqnum(
2329 direction_t direction
,
2335 return generic_array_bisect(f
,
2336 le64toh(f
->header
->entry_array_offset
),
2337 le64toh(f
->header
->n_entries
),
2344 static int test_object_realtime(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2351 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2355 if (le64toh(o
->entry
.realtime
) == needle
)
2357 else if (le64toh(o
->entry
.realtime
) < needle
)
2363 int journal_file_move_to_entry_by_realtime(
2366 direction_t direction
,
2372 return generic_array_bisect(f
,
2373 le64toh(f
->header
->entry_array_offset
),
2374 le64toh(f
->header
->n_entries
),
2376 test_object_realtime
,
2381 static int test_object_monotonic(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2388 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2392 if (le64toh(o
->entry
.monotonic
) == needle
)
2394 else if (le64toh(o
->entry
.monotonic
) < needle
)
2400 static int find_data_object_by_boot_id(
2406 char t
[sizeof("_BOOT_ID=")-1 + 32 + 1] = "_BOOT_ID=";
2408 sd_id128_to_string(boot_id
, t
+ 9);
2409 return journal_file_find_data_object(f
, t
, sizeof(t
) - 1, o
, b
);
2412 int journal_file_move_to_entry_by_monotonic(
2416 direction_t direction
,
2425 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, NULL
);
2431 return generic_array_bisect_plus_one(f
,
2432 le64toh(o
->data
.entry_offset
),
2433 le64toh(o
->data
.entry_array_offset
),
2434 le64toh(o
->data
.n_entries
),
2436 test_object_monotonic
,
2441 void journal_file_reset_location(JournalFile
*f
) {
2442 f
->location_type
= LOCATION_HEAD
;
2443 f
->current_offset
= 0;
2444 f
->current_seqnum
= 0;
2445 f
->current_realtime
= 0;
2446 f
->current_monotonic
= 0;
2447 zero(f
->current_boot_id
);
2448 f
->current_xor_hash
= 0;
2451 void journal_file_save_location(JournalFile
*f
, Object
*o
, uint64_t offset
) {
2452 f
->location_type
= LOCATION_SEEK
;
2453 f
->current_offset
= offset
;
2454 f
->current_seqnum
= le64toh(o
->entry
.seqnum
);
2455 f
->current_realtime
= le64toh(o
->entry
.realtime
);
2456 f
->current_monotonic
= le64toh(o
->entry
.monotonic
);
2457 f
->current_boot_id
= o
->entry
.boot_id
;
2458 f
->current_xor_hash
= le64toh(o
->entry
.xor_hash
);
2461 int journal_file_compare_locations(JournalFile
*af
, JournalFile
*bf
) {
2466 assert(af
->location_type
== LOCATION_SEEK
);
2467 assert(bf
->location_type
== LOCATION_SEEK
);
2469 /* If contents and timestamps match, these entries are
2470 * identical, even if the seqnum does not match */
2471 if (sd_id128_equal(af
->current_boot_id
, bf
->current_boot_id
) &&
2472 af
->current_monotonic
== bf
->current_monotonic
&&
2473 af
->current_realtime
== bf
->current_realtime
&&
2474 af
->current_xor_hash
== bf
->current_xor_hash
)
2477 if (sd_id128_equal(af
->header
->seqnum_id
, bf
->header
->seqnum_id
)) {
2479 /* If this is from the same seqnum source, compare
2481 if (af
->current_seqnum
< bf
->current_seqnum
)
2483 if (af
->current_seqnum
> bf
->current_seqnum
)
2486 /* Wow! This is weird, different data but the same
2487 * seqnums? Something is borked, but let's make the
2488 * best of it and compare by time. */
2491 if (sd_id128_equal(af
->current_boot_id
, bf
->current_boot_id
)) {
2493 /* If the boot id matches, compare monotonic time */
2494 if (af
->current_monotonic
< bf
->current_monotonic
)
2496 if (af
->current_monotonic
> bf
->current_monotonic
)
2500 /* Otherwise, compare UTC time */
2501 if (af
->current_realtime
< bf
->current_realtime
)
2503 if (af
->current_realtime
> bf
->current_realtime
)
2506 /* Finally, compare by contents */
2507 if (af
->current_xor_hash
< bf
->current_xor_hash
)
2509 if (af
->current_xor_hash
> bf
->current_xor_hash
)
2515 static int bump_array_index(uint64_t *i
, direction_t direction
, uint64_t n
) {
2517 /* Increase or decrease the specified index, in the right direction. */
2519 if (direction
== DIRECTION_DOWN
) {
2534 static bool check_properly_ordered(uint64_t new_offset
, uint64_t old_offset
, direction_t direction
) {
2536 /* Consider it an error if any of the two offsets is uninitialized */
2537 if (old_offset
== 0 || new_offset
== 0)
2540 /* If we go down, the new offset must be larger than the old one. */
2541 return direction
== DIRECTION_DOWN
?
2542 new_offset
> old_offset
:
2543 new_offset
< old_offset
;
2546 int journal_file_next_entry(
2549 direction_t direction
,
2550 Object
**ret
, uint64_t *offset
) {
2558 n
= le64toh(f
->header
->n_entries
);
2563 i
= direction
== DIRECTION_DOWN
? 0 : n
- 1;
2565 r
= generic_array_bisect(f
,
2566 le64toh(f
->header
->entry_array_offset
),
2567 le64toh(f
->header
->n_entries
),
2576 r
= bump_array_index(&i
, direction
, n
);
2581 /* And jump to it */
2583 r
= generic_array_get(f
,
2584 le64toh(f
->header
->entry_array_offset
),
2592 /* OK, so this entry is borked. Most likely some entry didn't get synced to disk properly, let's see if
2593 * the next one might work for us instead. */
2594 log_debug_errno(r
, "Entry item %" PRIu64
" is bad, skipping over it.", i
);
2596 r
= bump_array_index(&i
, direction
, n
);
2601 /* Ensure our array is properly ordered. */
2602 if (p
> 0 && !check_properly_ordered(ofs
, p
, direction
)) {
2603 log_debug("%s: entry array not properly ordered at entry %" PRIu64
, f
->path
, i
);
2613 int journal_file_next_entry_for_data(
2615 Object
*o
, uint64_t p
,
2616 uint64_t data_offset
,
2617 direction_t direction
,
2618 Object
**ret
, uint64_t *offset
) {
2625 assert(p
> 0 || !o
);
2627 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2631 n
= le64toh(d
->data
.n_entries
);
2636 i
= direction
== DIRECTION_DOWN
? 0 : n
- 1;
2638 if (o
->object
.type
!= OBJECT_ENTRY
)
2641 r
= generic_array_bisect_plus_one(f
,
2642 le64toh(d
->data
.entry_offset
),
2643 le64toh(d
->data
.entry_array_offset
),
2644 le64toh(d
->data
.n_entries
),
2654 r
= bump_array_index(&i
, direction
, n
);
2660 r
= generic_array_get_plus_one(f
,
2661 le64toh(d
->data
.entry_offset
),
2662 le64toh(d
->data
.entry_array_offset
),
2670 log_debug_errno(r
, "Data entry item %" PRIu64
" is bad, skipping over it.", i
);
2672 r
= bump_array_index(&i
, direction
, n
);
2677 /* Ensure our array is properly ordered. */
2678 if (p
> 0 && check_properly_ordered(ofs
, p
, direction
)) {
2679 log_debug("%s data entry array not properly ordered at entry %" PRIu64
, f
->path
, i
);
2689 int journal_file_move_to_entry_by_offset_for_data(
2691 uint64_t data_offset
,
2693 direction_t direction
,
2694 Object
**ret
, uint64_t *offset
) {
2701 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2705 return generic_array_bisect_plus_one(f
,
2706 le64toh(d
->data
.entry_offset
),
2707 le64toh(d
->data
.entry_array_offset
),
2708 le64toh(d
->data
.n_entries
),
2715 int journal_file_move_to_entry_by_monotonic_for_data(
2717 uint64_t data_offset
,
2720 direction_t direction
,
2721 Object
**ret
, uint64_t *offset
) {
2729 /* First, seek by time */
2730 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, &b
);
2736 r
= generic_array_bisect_plus_one(f
,
2737 le64toh(o
->data
.entry_offset
),
2738 le64toh(o
->data
.entry_array_offset
),
2739 le64toh(o
->data
.n_entries
),
2741 test_object_monotonic
,
2747 /* And now, continue seeking until we find an entry that
2748 * exists in both bisection arrays */
2754 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2758 r
= generic_array_bisect_plus_one(f
,
2759 le64toh(d
->data
.entry_offset
),
2760 le64toh(d
->data
.entry_array_offset
),
2761 le64toh(d
->data
.n_entries
),
2769 r
= journal_file_move_to_object(f
, OBJECT_DATA
, b
, &o
);
2773 r
= generic_array_bisect_plus_one(f
,
2774 le64toh(o
->data
.entry_offset
),
2775 le64toh(o
->data
.entry_array_offset
),
2776 le64toh(o
->data
.n_entries
),
2798 int journal_file_move_to_entry_by_seqnum_for_data(
2800 uint64_t data_offset
,
2802 direction_t direction
,
2803 Object
**ret
, uint64_t *offset
) {
2810 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2814 return generic_array_bisect_plus_one(f
,
2815 le64toh(d
->data
.entry_offset
),
2816 le64toh(d
->data
.entry_array_offset
),
2817 le64toh(d
->data
.n_entries
),
2824 int journal_file_move_to_entry_by_realtime_for_data(
2826 uint64_t data_offset
,
2828 direction_t direction
,
2829 Object
**ret
, uint64_t *offset
) {
2836 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2840 return generic_array_bisect_plus_one(f
,
2841 le64toh(d
->data
.entry_offset
),
2842 le64toh(d
->data
.entry_array_offset
),
2843 le64toh(d
->data
.n_entries
),
2845 test_object_realtime
,
2850 void journal_file_dump(JournalFile
*f
) {
2858 journal_file_print_header(f
);
2860 p
= le64toh(f
->header
->header_size
);
2862 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &o
);
2866 switch (o
->object
.type
) {
2869 printf("Type: OBJECT_UNUSED\n");
2873 printf("Type: OBJECT_DATA\n");
2877 printf("Type: OBJECT_FIELD\n");
2881 printf("Type: OBJECT_ENTRY seqnum=%"PRIu64
" monotonic=%"PRIu64
" realtime=%"PRIu64
"\n",
2882 le64toh(o
->entry
.seqnum
),
2883 le64toh(o
->entry
.monotonic
),
2884 le64toh(o
->entry
.realtime
));
2887 case OBJECT_FIELD_HASH_TABLE
:
2888 printf("Type: OBJECT_FIELD_HASH_TABLE\n");
2891 case OBJECT_DATA_HASH_TABLE
:
2892 printf("Type: OBJECT_DATA_HASH_TABLE\n");
2895 case OBJECT_ENTRY_ARRAY
:
2896 printf("Type: OBJECT_ENTRY_ARRAY\n");
2900 printf("Type: OBJECT_TAG seqnum=%"PRIu64
" epoch=%"PRIu64
"\n",
2901 le64toh(o
->tag
.seqnum
),
2902 le64toh(o
->tag
.epoch
));
2906 printf("Type: unknown (%i)\n", o
->object
.type
);
2910 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
)
2911 printf("Flags: %s\n",
2912 object_compressed_to_string(o
->object
.flags
& OBJECT_COMPRESSION_MASK
));
2914 if (p
== le64toh(f
->header
->tail_object_offset
))
2917 p
= p
+ ALIGN64(le64toh(o
->object
.size
));
2922 log_error("File corrupt");
2925 static const char* format_timestamp_safe(char *buf
, size_t l
, usec_t t
) {
2928 x
= format_timestamp(buf
, l
, t
);
2934 void journal_file_print_header(JournalFile
*f
) {
2935 char a
[33], b
[33], c
[33], d
[33];
2936 char x
[FORMAT_TIMESTAMP_MAX
], y
[FORMAT_TIMESTAMP_MAX
], z
[FORMAT_TIMESTAMP_MAX
];
2938 char bytes
[FORMAT_BYTES_MAX
];
2943 printf("File Path: %s\n"
2947 "Sequential Number ID: %s\n"
2949 "Compatible Flags:%s%s\n"
2950 "Incompatible Flags:%s%s%s\n"
2951 "Header size: %"PRIu64
"\n"
2952 "Arena size: %"PRIu64
"\n"
2953 "Data Hash Table Size: %"PRIu64
"\n"
2954 "Field Hash Table Size: %"PRIu64
"\n"
2955 "Rotate Suggested: %s\n"
2956 "Head Sequential Number: %"PRIu64
" (%"PRIx64
")\n"
2957 "Tail Sequential Number: %"PRIu64
" (%"PRIx64
")\n"
2958 "Head Realtime Timestamp: %s (%"PRIx64
")\n"
2959 "Tail Realtime Timestamp: %s (%"PRIx64
")\n"
2960 "Tail Monotonic Timestamp: %s (%"PRIx64
")\n"
2961 "Objects: %"PRIu64
"\n"
2962 "Entry Objects: %"PRIu64
"\n",
2964 sd_id128_to_string(f
->header
->file_id
, a
),
2965 sd_id128_to_string(f
->header
->machine_id
, b
),
2966 sd_id128_to_string(f
->header
->boot_id
, c
),
2967 sd_id128_to_string(f
->header
->seqnum_id
, d
),
2968 f
->header
->state
== STATE_OFFLINE
? "OFFLINE" :
2969 f
->header
->state
== STATE_ONLINE
? "ONLINE" :
2970 f
->header
->state
== STATE_ARCHIVED
? "ARCHIVED" : "UNKNOWN",
2971 JOURNAL_HEADER_SEALED(f
->header
) ? " SEALED" : "",
2972 (le32toh(f
->header
->compatible_flags
) & ~HEADER_COMPATIBLE_ANY
) ? " ???" : "",
2973 JOURNAL_HEADER_COMPRESSED_XZ(f
->header
) ? " COMPRESSED-XZ" : "",
2974 JOURNAL_HEADER_COMPRESSED_LZ4(f
->header
) ? " COMPRESSED-LZ4" : "",
2975 (le32toh(f
->header
->incompatible_flags
) & ~HEADER_INCOMPATIBLE_ANY
) ? " ???" : "",
2976 le64toh(f
->header
->header_size
),
2977 le64toh(f
->header
->arena_size
),
2978 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
2979 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
),
2980 yes_no(journal_file_rotate_suggested(f
, 0)),
2981 le64toh(f
->header
->head_entry_seqnum
), le64toh(f
->header
->head_entry_seqnum
),
2982 le64toh(f
->header
->tail_entry_seqnum
), le64toh(f
->header
->tail_entry_seqnum
),
2983 format_timestamp_safe(x
, sizeof(x
), le64toh(f
->header
->head_entry_realtime
)), le64toh(f
->header
->head_entry_realtime
),
2984 format_timestamp_safe(y
, sizeof(y
), le64toh(f
->header
->tail_entry_realtime
)), le64toh(f
->header
->tail_entry_realtime
),
2985 format_timespan(z
, sizeof(z
), le64toh(f
->header
->tail_entry_monotonic
), USEC_PER_MSEC
), le64toh(f
->header
->tail_entry_monotonic
),
2986 le64toh(f
->header
->n_objects
),
2987 le64toh(f
->header
->n_entries
));
2989 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
2990 printf("Data Objects: %"PRIu64
"\n"
2991 "Data Hash Table Fill: %.1f%%\n",
2992 le64toh(f
->header
->n_data
),
2993 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))));
2995 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
2996 printf("Field Objects: %"PRIu64
"\n"
2997 "Field Hash Table Fill: %.1f%%\n",
2998 le64toh(f
->header
->n_fields
),
2999 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))));
3001 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_tags
))
3002 printf("Tag Objects: %"PRIu64
"\n",
3003 le64toh(f
->header
->n_tags
));
3004 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
3005 printf("Entry Array Objects: %"PRIu64
"\n",
3006 le64toh(f
->header
->n_entry_arrays
));
3008 if (fstat(f
->fd
, &st
) >= 0)
3009 printf("Disk usage: %s\n", format_bytes(bytes
, sizeof(bytes
), (uint64_t) st
.st_blocks
* 512ULL));
3012 static int journal_file_warn_btrfs(JournalFile
*f
) {
3018 /* Before we write anything, check if the COW logic is turned
3019 * off on btrfs. Given our write pattern that is quite
3020 * unfriendly to COW file systems this should greatly improve
3021 * performance on COW file systems, such as btrfs, at the
3022 * expense of data integrity features (which shouldn't be too
3023 * bad, given that we do our own checksumming). */
3025 r
= btrfs_is_filesystem(f
->fd
);
3027 return log_warning_errno(r
, "Failed to determine if journal is on btrfs: %m");
3031 r
= read_attr_fd(f
->fd
, &attrs
);
3033 return log_warning_errno(r
, "Failed to read file attributes: %m");
3035 if (attrs
& FS_NOCOW_FL
) {
3036 log_debug("Detected btrfs file system with copy-on-write disabled, all is good.");
3040 log_notice("Creating journal file %s on a btrfs file system, and copy-on-write is enabled. "
3041 "This is likely to slow down journal access substantially, please consider turning "
3042 "off the copy-on-write file attribute on the journal directory, using chattr +C.", f
->path
);
3047 int journal_file_open(
3054 JournalMetrics
*metrics
,
3055 MMapCache
*mmap_cache
,
3056 Set
*deferred_closes
,
3057 JournalFile
*template,
3058 JournalFile
**ret
) {
3060 bool newly_created
= false;
3066 assert(fd
>= 0 || fname
);
3068 if ((flags
& O_ACCMODE
) != O_RDONLY
&&
3069 (flags
& O_ACCMODE
) != O_RDWR
)
3073 if (!endswith(fname
, ".journal") &&
3074 !endswith(fname
, ".journal~"))
3078 f
= new0(JournalFile
, 1);
3086 f
->prot
= prot_from_flags(flags
);
3087 f
->writable
= (flags
& O_ACCMODE
) != O_RDONLY
;
3088 #if defined(HAVE_LZ4)
3089 f
->compress_lz4
= compress
;
3090 #elif defined(HAVE_XZ)
3091 f
->compress_xz
= compress
;
3098 f
->mmap
= mmap_cache_ref(mmap_cache
);
3100 f
->mmap
= mmap_cache_new();
3108 f
->path
= strdup(fname
);
3114 /* If we don't know the path, fill in something explanatory and vaguely useful */
3115 if (asprintf(&f
->path
, "/proc/self/%i", fd
) < 0) {
3121 f
->chain_cache
= ordered_hashmap_new(&uint64_hash_ops
);
3122 if (!f
->chain_cache
) {
3128 f
->fd
= open(f
->path
, f
->flags
|O_CLOEXEC
, f
->mode
);
3134 /* fds we opened here by us should also be closed by us. */
3138 r
= journal_file_fstat(f
);
3142 if (f
->last_stat
.st_size
== 0 && f
->writable
) {
3144 (void) journal_file_warn_btrfs(f
);
3146 /* Let's attach the creation time to the journal file,
3147 * so that the vacuuming code knows the age of this
3148 * file even if the file might end up corrupted one
3149 * day... Ideally we'd just use the creation time many
3150 * file systems maintain for each file, but there is
3151 * currently no usable API to query this, hence let's
3152 * emulate this via extended attributes. If extended
3153 * attributes are not supported we'll just skip this,
3154 * and rely solely on mtime/atime/ctime of the file. */
3156 fd_setcrtime(f
->fd
, 0);
3159 /* Try to load the FSPRG state, and if we can't, then
3160 * just don't do sealing */
3162 r
= journal_file_fss_load(f
);
3168 r
= journal_file_init_header(f
, template);
3172 r
= journal_file_fstat(f
);
3176 newly_created
= true;
3179 if (f
->last_stat
.st_size
< (off_t
) HEADER_SIZE_MIN
) {
3184 r
= mmap_cache_get(f
->mmap
, f
->fd
, f
->prot
, CONTEXT_HEADER
, true, 0, PAGE_ALIGN(sizeof(Header
)), &f
->last_stat
, &h
);
3190 if (!newly_created
) {
3191 if (deferred_closes
)
3192 journal_file_close_set(deferred_closes
);
3194 r
= journal_file_verify_header(f
);
3200 if (!newly_created
&& f
->writable
) {
3201 r
= journal_file_fss_load(f
);
3209 journal_default_metrics(metrics
, f
->fd
);
3210 f
->metrics
= *metrics
;
3211 } else if (template)
3212 f
->metrics
= template->metrics
;
3214 r
= journal_file_refresh_header(f
);
3220 r
= journal_file_hmac_setup(f
);
3225 if (newly_created
) {
3226 r
= journal_file_setup_field_hash_table(f
);
3230 r
= journal_file_setup_data_hash_table(f
);
3235 r
= journal_file_append_first_tag(f
);
3241 if (mmap_cache_got_sigbus(f
->mmap
, f
->fd
)) {
3246 if (template && template->post_change_timer
) {
3247 r
= journal_file_enable_post_change_timer(
3249 sd_event_source_get_event(template->post_change_timer
),
3250 template->post_change_timer_period
);
3256 /* The file is opened now successfully, thus we take possession of any passed in fd. */
3263 if (f
->fd
>= 0 && mmap_cache_got_sigbus(f
->mmap
, f
->fd
))
3266 (void) journal_file_close(f
);
3271 int journal_file_rotate(JournalFile
**f
, bool compress
, bool seal
, Set
*deferred_closes
) {
3272 _cleanup_free_
char *p
= NULL
;
3274 JournalFile
*old_file
, *new_file
= NULL
;
3282 if (!old_file
->writable
)
3285 /* Is this a journal file that was passed to us as fd? If so, we synthesized a path name for it, and we refuse
3286 * rotation, since we don't know the actual path, and couldn't rename the file hence. */
3287 if (path_startswith(old_file
->path
, "/proc/self/fd"))
3290 if (!endswith(old_file
->path
, ".journal"))
3293 l
= strlen(old_file
->path
);
3294 r
= asprintf(&p
, "%.*s@" SD_ID128_FORMAT_STR
"-%016"PRIx64
"-%016"PRIx64
".journal",
3295 (int) l
- 8, old_file
->path
,
3296 SD_ID128_FORMAT_VAL(old_file
->header
->seqnum_id
),
3297 le64toh((*f
)->header
->head_entry_seqnum
),
3298 le64toh((*f
)->header
->head_entry_realtime
));
3302 /* Try to rename the file to the archived version. If the file
3303 * already was deleted, we'll get ENOENT, let's ignore that
3305 r
= rename(old_file
->path
, p
);
3306 if (r
< 0 && errno
!= ENOENT
)
3309 /* Sync the rename to disk */
3310 (void) fsync_directory_of_file(old_file
->fd
);
3312 /* Set as archive so offlining commits w/state=STATE_ARCHIVED.
3313 * Previously we would set old_file->header->state to STATE_ARCHIVED directly here,
3314 * but journal_file_set_offline() short-circuits when state != STATE_ONLINE, which
3315 * would result in the rotated journal never getting fsync() called before closing.
3316 * Now we simply queue the archive state by setting an archive bit, leaving the state
3317 * as STATE_ONLINE so proper offlining occurs. */
3318 old_file
->archive
= true;
3320 /* Currently, btrfs is not very good with out write patterns
3321 * and fragments heavily. Let's defrag our journal files when
3322 * we archive them */
3323 old_file
->defrag_on_close
= true;
3325 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
);
3327 if (deferred_closes
&&
3328 set_put(deferred_closes
, old_file
) >= 0)
3329 (void) journal_file_set_offline(old_file
, false);
3331 (void) journal_file_close(old_file
);
3337 int journal_file_open_reliably(
3343 JournalMetrics
*metrics
,
3344 MMapCache
*mmap_cache
,
3345 Set
*deferred_closes
,
3346 JournalFile
*template,
3347 JournalFile
**ret
) {
3351 _cleanup_free_
char *p
= NULL
;
3353 r
= journal_file_open(-1, fname
, flags
, mode
, compress
, seal
, metrics
, mmap_cache
, deferred_closes
, template, ret
);
3355 -EBADMSG
, /* Corrupted */
3356 -ENODATA
, /* Truncated */
3357 -EHOSTDOWN
, /* Other machine */
3358 -EPROTONOSUPPORT
, /* Incompatible feature */
3359 -EBUSY
, /* Unclean shutdown */
3360 -ESHUTDOWN
, /* Already archived */
3361 -EIO
, /* IO error, including SIGBUS on mmap */
3362 -EIDRM
, /* File has been deleted */
3363 -ETXTBSY
)) /* File is from the future */
3366 if ((flags
& O_ACCMODE
) == O_RDONLY
)
3369 if (!(flags
& O_CREAT
))
3372 if (!endswith(fname
, ".journal"))
3375 /* The file is corrupted. Rotate it away and try it again (but only once) */
3378 if (asprintf(&p
, "%.*s@%016"PRIx64
"-%016"PRIx64
".journal~",
3380 now(CLOCK_REALTIME
),
3384 if (rename(fname
, p
) < 0)
3387 /* btrfs doesn't cope well with our write pattern and
3388 * fragments heavily. Let's defrag all files we rotate */
3390 (void) chattr_path(p
, 0, FS_NOCOW_FL
);
3391 (void) btrfs_defrag(p
);
3393 log_warning_errno(r
, "File %s corrupted or uncleanly shut down, renaming and replacing.", fname
);
3395 return journal_file_open(-1, fname
, flags
, mode
, compress
, seal
, metrics
, mmap_cache
, deferred_closes
, template, ret
);
3398 int journal_file_copy_entry(JournalFile
*from
, JournalFile
*to
, Object
*o
, uint64_t p
, uint64_t *seqnum
, Object
**ret
, uint64_t *offset
) {
3400 uint64_t q
, xor_hash
= 0;
3413 ts
.monotonic
= le64toh(o
->entry
.monotonic
);
3414 ts
.realtime
= le64toh(o
->entry
.realtime
);
3416 n
= journal_file_entry_n_items(o
);
3417 /* alloca() can't take 0, hence let's allocate at least one */
3418 items
= alloca(sizeof(EntryItem
) * MAX(1u, n
));
3420 for (i
= 0; i
< n
; i
++) {
3427 q
= le64toh(o
->entry
.items
[i
].object_offset
);
3428 le_hash
= o
->entry
.items
[i
].hash
;
3430 r
= journal_file_move_to_object(from
, OBJECT_DATA
, q
, &o
);
3434 if (le_hash
!= o
->data
.hash
)
3437 l
= le64toh(o
->object
.size
) - offsetof(Object
, data
.payload
);
3440 /* We hit the limit on 32bit machines */
3441 if ((uint64_t) t
!= l
)
3444 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
) {
3445 #if defined(HAVE_XZ) || defined(HAVE_LZ4)
3448 r
= decompress_blob(o
->object
.flags
& OBJECT_COMPRESSION_MASK
,
3449 o
->data
.payload
, l
, &from
->compress_buffer
, &from
->compress_buffer_size
, &rsize
, 0);
3453 data
= from
->compress_buffer
;
3456 return -EPROTONOSUPPORT
;
3459 data
= o
->data
.payload
;
3461 r
= journal_file_append_data(to
, data
, l
, &u
, &h
);
3465 xor_hash
^= le64toh(u
->data
.hash
);
3466 items
[i
].object_offset
= htole64(h
);
3467 items
[i
].hash
= u
->data
.hash
;
3469 r
= journal_file_move_to_object(from
, OBJECT_ENTRY
, p
, &o
);
3474 r
= journal_file_append_entry_internal(to
, &ts
, xor_hash
, items
, n
, seqnum
, ret
, offset
);
3476 if (mmap_cache_got_sigbus(to
->mmap
, to
->fd
))
3482 void journal_reset_metrics(JournalMetrics
*m
) {
3485 /* Set everything to "pick automatic values". */
3487 *m
= (JournalMetrics
) {
3488 .min_use
= (uint64_t) -1,
3489 .max_use
= (uint64_t) -1,
3490 .min_size
= (uint64_t) -1,
3491 .max_size
= (uint64_t) -1,
3492 .keep_free
= (uint64_t) -1,
3493 .n_max_files
= (uint64_t) -1,
3497 void journal_default_metrics(JournalMetrics
*m
, int fd
) {
3498 char a
[FORMAT_BYTES_MAX
], b
[FORMAT_BYTES_MAX
], c
[FORMAT_BYTES_MAX
], d
[FORMAT_BYTES_MAX
], e
[FORMAT_BYTES_MAX
];
3505 if (fstatvfs(fd
, &ss
) >= 0)
3506 fs_size
= ss
.f_frsize
* ss
.f_blocks
;
3508 log_debug_errno(errno
, "Failed to detremine disk size: %m");
3512 if (m
->max_use
== (uint64_t) -1) {
3515 m
->max_use
= PAGE_ALIGN(fs_size
/ 10); /* 10% of file system size */
3517 if (m
->max_use
> DEFAULT_MAX_USE_UPPER
)
3518 m
->max_use
= DEFAULT_MAX_USE_UPPER
;
3520 if (m
->max_use
< DEFAULT_MAX_USE_LOWER
)
3521 m
->max_use
= DEFAULT_MAX_USE_LOWER
;
3523 m
->max_use
= DEFAULT_MAX_USE_LOWER
;
3525 m
->max_use
= PAGE_ALIGN(m
->max_use
);
3527 if (m
->max_use
!= 0 && m
->max_use
< JOURNAL_FILE_SIZE_MIN
*2)
3528 m
->max_use
= JOURNAL_FILE_SIZE_MIN
*2;
3531 if (m
->min_use
== (uint64_t) -1)
3532 m
->min_use
= DEFAULT_MIN_USE
;
3534 if (m
->min_use
> m
->max_use
)
3535 m
->min_use
= m
->max_use
;
3537 if (m
->max_size
== (uint64_t) -1) {
3538 m
->max_size
= PAGE_ALIGN(m
->max_use
/ 8); /* 8 chunks */
3540 if (m
->max_size
> DEFAULT_MAX_SIZE_UPPER
)
3541 m
->max_size
= DEFAULT_MAX_SIZE_UPPER
;
3543 m
->max_size
= PAGE_ALIGN(m
->max_size
);
3545 if (m
->max_size
!= 0) {
3546 if (m
->max_size
< JOURNAL_FILE_SIZE_MIN
)
3547 m
->max_size
= JOURNAL_FILE_SIZE_MIN
;
3549 if (m
->max_use
!= 0 && m
->max_size
*2 > m
->max_use
)
3550 m
->max_use
= m
->max_size
*2;
3553 if (m
->min_size
== (uint64_t) -1)
3554 m
->min_size
= JOURNAL_FILE_SIZE_MIN
;
3556 m
->min_size
= PAGE_ALIGN(m
->min_size
);
3558 if (m
->min_size
< JOURNAL_FILE_SIZE_MIN
)
3559 m
->min_size
= JOURNAL_FILE_SIZE_MIN
;
3561 if (m
->max_size
!= 0 && m
->min_size
> m
->max_size
)
3562 m
->max_size
= m
->min_size
;
3565 if (m
->keep_free
== (uint64_t) -1) {
3568 m
->keep_free
= PAGE_ALIGN(fs_size
* 3 / 20); /* 15% of file system size */
3570 if (m
->keep_free
> DEFAULT_KEEP_FREE_UPPER
)
3571 m
->keep_free
= DEFAULT_KEEP_FREE_UPPER
;
3574 m
->keep_free
= DEFAULT_KEEP_FREE
;
3577 if (m
->n_max_files
== (uint64_t) -1)
3578 m
->n_max_files
= DEFAULT_N_MAX_FILES
;
3580 log_debug("Fixed min_use=%s max_use=%s max_size=%s min_size=%s keep_free=%s n_max_files=%" PRIu64
,
3581 format_bytes(a
, sizeof(a
), m
->min_use
),
3582 format_bytes(b
, sizeof(b
), m
->max_use
),
3583 format_bytes(c
, sizeof(c
), m
->max_size
),
3584 format_bytes(d
, sizeof(d
), m
->min_size
),
3585 format_bytes(e
, sizeof(e
), m
->keep_free
),
3589 int journal_file_get_cutoff_realtime_usec(JournalFile
*f
, usec_t
*from
, usec_t
*to
) {
3595 if (f
->header
->head_entry_realtime
== 0)
3598 *from
= le64toh(f
->header
->head_entry_realtime
);
3602 if (f
->header
->tail_entry_realtime
== 0)
3605 *to
= le64toh(f
->header
->tail_entry_realtime
);
3611 int journal_file_get_cutoff_monotonic_usec(JournalFile
*f
, sd_id128_t boot_id
, usec_t
*from
, usec_t
*to
) {
3619 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, &p
);
3623 if (le64toh(o
->data
.n_entries
) <= 0)
3627 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, le64toh(o
->data
.entry_offset
), &o
);
3631 *from
= le64toh(o
->entry
.monotonic
);
3635 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
3639 r
= generic_array_get_plus_one(f
,
3640 le64toh(o
->data
.entry_offset
),
3641 le64toh(o
->data
.entry_array_offset
),
3642 le64toh(o
->data
.n_entries
)-1,
3647 *to
= le64toh(o
->entry
.monotonic
);
3653 bool journal_file_rotate_suggested(JournalFile
*f
, usec_t max_file_usec
) {
3657 /* If we gained new header fields we gained new features,
3658 * hence suggest a rotation */
3659 if (le64toh(f
->header
->header_size
) < sizeof(Header
)) {
3660 log_debug("%s uses an outdated header, suggesting rotation.", f
->path
);
3664 /* Let's check if the hash tables grew over a certain fill
3665 * level (75%, borrowing this value from Java's hash table
3666 * implementation), and if so suggest a rotation. To calculate
3667 * the fill level we need the n_data field, which only exists
3668 * in newer versions. */
3670 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
3671 if (le64toh(f
->header
->n_data
) * 4ULL > (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
3672 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.",
3674 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))),
3675 le64toh(f
->header
->n_data
),
3676 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
3677 (unsigned long long) f
->last_stat
.st_size
,
3678 f
->last_stat
.st_size
/ le64toh(f
->header
->n_data
));
3682 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
3683 if (le64toh(f
->header
->n_fields
) * 4ULL > (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
3684 log_debug("Field hash table of %s has a fill level at %.1f (%"PRIu64
" of %"PRIu64
" items), suggesting rotation.",
3686 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))),
3687 le64toh(f
->header
->n_fields
),
3688 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
));
3692 /* Are the data objects properly indexed by field objects? */
3693 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
) &&
3694 JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
) &&
3695 le64toh(f
->header
->n_data
) > 0 &&
3696 le64toh(f
->header
->n_fields
) == 0)
3699 if (max_file_usec
> 0) {
3702 h
= le64toh(f
->header
->head_entry_realtime
);
3703 t
= now(CLOCK_REALTIME
);
3705 if (h
> 0 && t
> h
+ max_file_usec
)