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"
45 #include "xattr-util.h"
47 #define DEFAULT_DATA_HASH_TABLE_SIZE (2047ULL*sizeof(HashItem))
48 #define DEFAULT_FIELD_HASH_TABLE_SIZE (333ULL*sizeof(HashItem))
50 #define COMPRESSION_SIZE_THRESHOLD (512ULL)
52 /* This is the minimum journal file size */
53 #define JOURNAL_FILE_SIZE_MIN (512ULL*1024ULL) /* 512 KiB */
55 /* These are the lower and upper bounds if we deduce the max_use value
56 * from the file system size */
57 #define DEFAULT_MAX_USE_LOWER (1ULL*1024ULL*1024ULL) /* 1 MiB */
58 #define DEFAULT_MAX_USE_UPPER (4ULL*1024ULL*1024ULL*1024ULL) /* 4 GiB */
60 /* This is the default minimal use limit, how much we'll use even if keep_free suggests otherwise. */
61 #define DEFAULT_MIN_USE (1ULL*1024ULL*1024ULL) /* 1 MiB */
63 /* This is the upper bound if we deduce max_size from max_use */
64 #define DEFAULT_MAX_SIZE_UPPER (128ULL*1024ULL*1024ULL) /* 128 MiB */
66 /* This is the upper bound if we deduce the keep_free value from the
68 #define DEFAULT_KEEP_FREE_UPPER (4ULL*1024ULL*1024ULL*1024ULL) /* 4 GiB */
70 /* This is the keep_free value when we can't determine the system
72 #define DEFAULT_KEEP_FREE (1024ULL*1024ULL) /* 1 MB */
74 /* This is the default maximum number of journal files to keep around. */
75 #define DEFAULT_N_MAX_FILES (100)
77 /* n_data was the first entry we added after the initial file format design */
78 #define HEADER_SIZE_MIN ALIGN64(offsetof(Header, n_data))
80 /* How many entries to keep in the entry array chain cache at max */
81 #define CHAIN_CACHE_MAX 20
83 /* How much to increase the journal file size at once each time we allocate something new. */
84 #define FILE_SIZE_INCREASE (8ULL*1024ULL*1024ULL) /* 8MB */
86 /* Reread fstat() of the file for detecting deletions at least this often */
87 #define LAST_STAT_REFRESH_USEC (5*USEC_PER_SEC)
89 /* The mmap context to use for the header we pick as one above the last defined typed */
90 #define CONTEXT_HEADER _OBJECT_TYPE_MAX
92 /* This may be called from a separate thread to prevent blocking the caller for the duration of fsync().
93 * As a result we use atomic operations on f->offline_state for inter-thread communications with
94 * journal_file_set_offline() and journal_file_set_online(). */
95 static void journal_file_set_offline_internal(JournalFile
*f
) {
101 switch (f
->offline_state
) {
103 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_CANCEL
, OFFLINE_DONE
))
107 case OFFLINE_AGAIN_FROM_SYNCING
:
108 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_AGAIN_FROM_SYNCING
, OFFLINE_SYNCING
))
112 case OFFLINE_AGAIN_FROM_OFFLINING
:
113 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_AGAIN_FROM_OFFLINING
, OFFLINE_SYNCING
))
117 case OFFLINE_SYNCING
:
120 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_SYNCING
, OFFLINE_OFFLINING
))
123 f
->header
->state
= f
->archive
? STATE_ARCHIVED
: STATE_OFFLINE
;
127 case OFFLINE_OFFLINING
:
128 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_OFFLINING
, OFFLINE_DONE
))
136 log_debug("OFFLINE_JOINED unexpected offline state for journal_file_set_offline_internal()");
142 static void * journal_file_set_offline_thread(void *arg
) {
143 JournalFile
*f
= arg
;
145 journal_file_set_offline_internal(f
);
150 static int journal_file_set_offline_thread_join(JournalFile
*f
) {
155 if (f
->offline_state
== OFFLINE_JOINED
)
158 r
= pthread_join(f
->offline_thread
, NULL
);
162 f
->offline_state
= OFFLINE_JOINED
;
164 if (mmap_cache_got_sigbus(f
->mmap
, f
->fd
))
170 /* Trigger a restart if the offline thread is mid-flight in a restartable state. */
171 static bool journal_file_set_offline_try_restart(JournalFile
*f
) {
173 switch (f
->offline_state
) {
174 case OFFLINE_AGAIN_FROM_SYNCING
:
175 case OFFLINE_AGAIN_FROM_OFFLINING
:
179 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_CANCEL
, OFFLINE_AGAIN_FROM_SYNCING
))
183 case OFFLINE_SYNCING
:
184 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_SYNCING
, OFFLINE_AGAIN_FROM_SYNCING
))
188 case OFFLINE_OFFLINING
:
189 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_OFFLINING
, OFFLINE_AGAIN_FROM_OFFLINING
))
199 /* Sets a journal offline.
201 * If wait is false then an offline is dispatched in a separate thread for a
202 * subsequent journal_file_set_offline() or journal_file_set_online() of the
203 * same journal to synchronize with.
205 * If wait is true, then either an existing offline thread will be restarted
206 * and joined, or if none exists the offline is simply performed in this
207 * context without involving another thread.
209 int journal_file_set_offline(JournalFile
*f
, bool wait
) {
218 if (!(f
->fd
>= 0 && f
->header
))
221 /* An offlining journal is implicitly online and may modify f->header->state,
222 * we must also join any potentially lingering offline thread when not online. */
223 if (!journal_file_is_offlining(f
) && f
->header
->state
!= STATE_ONLINE
)
224 return journal_file_set_offline_thread_join(f
);
226 /* Restart an in-flight offline thread and wait if needed, or join a lingering done one. */
227 restarted
= journal_file_set_offline_try_restart(f
);
228 if ((restarted
&& wait
) || !restarted
) {
229 r
= journal_file_set_offline_thread_join(f
);
237 /* Initiate a new offline. */
238 f
->offline_state
= OFFLINE_SYNCING
;
240 if (wait
) /* Without using a thread if waiting. */
241 journal_file_set_offline_internal(f
);
243 r
= pthread_create(&f
->offline_thread
, NULL
, journal_file_set_offline_thread
, f
);
245 f
->offline_state
= OFFLINE_JOINED
;
253 static int journal_file_set_online(JournalFile
*f
) {
261 if (!(f
->fd
>= 0 && f
->header
))
265 switch (f
->offline_state
) {
267 /* No offline thread, no need to wait. */
271 case OFFLINE_SYNCING
:
272 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_SYNCING
, OFFLINE_CANCEL
))
274 /* Canceled syncing prior to offlining, no need to wait. */
277 case OFFLINE_AGAIN_FROM_SYNCING
:
278 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_AGAIN_FROM_SYNCING
, OFFLINE_CANCEL
))
280 /* Canceled restart from syncing, no need to wait. */
283 case OFFLINE_AGAIN_FROM_OFFLINING
:
284 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_AGAIN_FROM_OFFLINING
, OFFLINE_CANCEL
))
286 /* Canceled restart from offlining, must wait for offlining to complete however. */
288 /* fall through to wait */
292 r
= journal_file_set_offline_thread_join(f
);
302 if (mmap_cache_got_sigbus(f
->mmap
, f
->fd
))
305 switch (f
->header
->state
) {
310 f
->header
->state
= STATE_ONLINE
;
319 bool journal_file_is_offlining(JournalFile
*f
) {
322 __sync_synchronize();
324 if (f
->offline_state
== OFFLINE_DONE
||
325 f
->offline_state
== OFFLINE_JOINED
)
331 JournalFile
* journal_file_close(JournalFile
*f
) {
335 /* Write the final tag */
336 if (f
->seal
&& f
->writable
) {
339 r
= journal_file_append_tag(f
);
341 log_error_errno(r
, "Failed to append tag when closing journal: %m");
345 if (f
->post_change_timer
) {
348 if (sd_event_source_get_enabled(f
->post_change_timer
, &enabled
) >= 0)
349 if (enabled
== SD_EVENT_ONESHOT
)
350 journal_file_post_change(f
);
352 (void) sd_event_source_set_enabled(f
->post_change_timer
, SD_EVENT_OFF
);
353 sd_event_source_unref(f
->post_change_timer
);
356 journal_file_set_offline(f
, true);
358 if (f
->mmap
&& f
->fd
>= 0)
359 mmap_cache_close_fd(f
->mmap
, f
->fd
);
361 if (f
->fd
>= 0 && f
->defrag_on_close
) {
363 /* Be friendly to btrfs: turn COW back on again now,
364 * and defragment the file. We won't write to the file
365 * ever again, hence remove all fragmentation, and
366 * reenable all the good bits COW usually provides
367 * (such as data checksumming). */
369 (void) chattr_fd(f
->fd
, 0, FS_NOCOW_FL
);
370 (void) btrfs_defrag_fd(f
->fd
);
377 mmap_cache_unref(f
->mmap
);
379 ordered_hashmap_free_free(f
->chain_cache
);
381 #if defined(HAVE_XZ) || defined(HAVE_LZ4)
382 free(f
->compress_buffer
);
387 munmap(f
->fss_file
, PAGE_ALIGN(f
->fss_file_size
));
389 free(f
->fsprg_state
);
394 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 int journal_file_verify_header(JournalFile
*f
) {
517 if (memcmp(f
->header
->signature
, HEADER_SIGNATURE
, 8))
520 /* In both read and write mode we refuse to open files with
521 * incompatible flags we don't know */
522 flags
= le32toh(f
->header
->incompatible_flags
);
523 if (flags
& ~HEADER_INCOMPATIBLE_SUPPORTED
) {
524 if (flags
& ~HEADER_INCOMPATIBLE_ANY
)
525 log_debug("Journal file %s has unknown incompatible flags %"PRIx32
,
526 f
->path
, flags
& ~HEADER_INCOMPATIBLE_ANY
);
527 flags
= (flags
& HEADER_INCOMPATIBLE_ANY
) & ~HEADER_INCOMPATIBLE_SUPPORTED
;
529 log_debug("Journal file %s uses incompatible flags %"PRIx32
530 " disabled at compilation time.", f
->path
, flags
);
531 return -EPROTONOSUPPORT
;
534 /* When open for writing we refuse to open files with
535 * compatible flags, too */
536 flags
= le32toh(f
->header
->compatible_flags
);
537 if (f
->writable
&& (flags
& ~HEADER_COMPATIBLE_SUPPORTED
)) {
538 if (flags
& ~HEADER_COMPATIBLE_ANY
)
539 log_debug("Journal file %s has unknown compatible flags %"PRIx32
,
540 f
->path
, flags
& ~HEADER_COMPATIBLE_ANY
);
541 flags
= (flags
& HEADER_COMPATIBLE_ANY
) & ~HEADER_COMPATIBLE_SUPPORTED
;
543 log_debug("Journal file %s uses compatible flags %"PRIx32
544 " disabled at compilation time.", f
->path
, flags
);
545 return -EPROTONOSUPPORT
;
548 if (f
->header
->state
>= _STATE_MAX
)
551 /* The first addition was n_data, so check that we are at least this large */
552 if (le64toh(f
->header
->header_size
) < HEADER_SIZE_MIN
)
555 if (JOURNAL_HEADER_SEALED(f
->header
) && !JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
558 if ((le64toh(f
->header
->header_size
) + le64toh(f
->header
->arena_size
)) > (uint64_t) f
->last_stat
.st_size
)
561 if (le64toh(f
->header
->tail_object_offset
) > (le64toh(f
->header
->header_size
) + le64toh(f
->header
->arena_size
)))
564 if (!VALID64(le64toh(f
->header
->data_hash_table_offset
)) ||
565 !VALID64(le64toh(f
->header
->field_hash_table_offset
)) ||
566 !VALID64(le64toh(f
->header
->tail_object_offset
)) ||
567 !VALID64(le64toh(f
->header
->entry_array_offset
)))
572 sd_id128_t machine_id
;
575 r
= sd_id128_get_machine(&machine_id
);
579 if (!sd_id128_equal(machine_id
, f
->header
->machine_id
))
582 state
= f
->header
->state
;
584 if (state
== STATE_ONLINE
) {
585 log_debug("Journal file %s is already online. Assuming unclean closing.", f
->path
);
587 } else if (state
== STATE_ARCHIVED
)
589 else if (state
!= STATE_OFFLINE
) {
590 log_debug("Journal file %s has unknown state %i.", f
->path
, state
);
595 f
->compress_xz
= JOURNAL_HEADER_COMPRESSED_XZ(f
->header
);
596 f
->compress_lz4
= JOURNAL_HEADER_COMPRESSED_LZ4(f
->header
);
598 f
->seal
= JOURNAL_HEADER_SEALED(f
->header
);
603 static int journal_file_fstat(JournalFile
*f
) {
607 if (fstat(f
->fd
, &f
->last_stat
) < 0)
610 f
->last_stat_usec
= now(CLOCK_MONOTONIC
);
612 /* Refuse appending to files that are already deleted */
613 if (f
->last_stat
.st_nlink
<= 0)
619 static int journal_file_allocate(JournalFile
*f
, uint64_t offset
, uint64_t size
) {
620 uint64_t old_size
, new_size
;
626 /* We assume that this file is not sparse, and we know that
627 * for sure, since we always call posix_fallocate()
630 if (mmap_cache_got_sigbus(f
->mmap
, f
->fd
))
634 le64toh(f
->header
->header_size
) +
635 le64toh(f
->header
->arena_size
);
637 new_size
= PAGE_ALIGN(offset
+ size
);
638 if (new_size
< le64toh(f
->header
->header_size
))
639 new_size
= le64toh(f
->header
->header_size
);
641 if (new_size
<= old_size
) {
643 /* We already pre-allocated enough space, but before
644 * we write to it, let's check with fstat() if the
645 * file got deleted, in order make sure we don't throw
646 * away the data immediately. Don't check fstat() for
647 * all writes though, but only once ever 10s. */
649 if (f
->last_stat_usec
+ LAST_STAT_REFRESH_USEC
> now(CLOCK_MONOTONIC
))
652 return journal_file_fstat(f
);
655 /* Allocate more space. */
657 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
660 if (new_size
> f
->metrics
.min_size
&& f
->metrics
.keep_free
> 0) {
663 if (fstatvfs(f
->fd
, &svfs
) >= 0) {
666 available
= LESS_BY((uint64_t) svfs
.f_bfree
* (uint64_t) svfs
.f_bsize
, f
->metrics
.keep_free
);
668 if (new_size
- old_size
> available
)
673 /* Increase by larger blocks at once */
674 new_size
= ((new_size
+FILE_SIZE_INCREASE
-1) / FILE_SIZE_INCREASE
) * FILE_SIZE_INCREASE
;
675 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
676 new_size
= f
->metrics
.max_size
;
678 /* Note that the glibc fallocate() fallback is very
679 inefficient, hence we try to minimize the allocation area
681 r
= posix_fallocate(f
->fd
, old_size
, new_size
- old_size
);
685 f
->header
->arena_size
= htole64(new_size
- le64toh(f
->header
->header_size
));
687 return journal_file_fstat(f
);
690 static unsigned type_to_context(ObjectType type
) {
691 /* One context for each type, plus one catch-all for the rest */
692 assert_cc(_OBJECT_TYPE_MAX
<= MMAP_CACHE_MAX_CONTEXTS
);
693 assert_cc(CONTEXT_HEADER
< MMAP_CACHE_MAX_CONTEXTS
);
694 return type
> OBJECT_UNUSED
&& type
< _OBJECT_TYPE_MAX
? type
: 0;
697 static int journal_file_move_to(JournalFile
*f
, ObjectType type
, bool keep_always
, uint64_t offset
, uint64_t size
, void **ret
) {
706 /* Avoid SIGBUS on invalid accesses */
707 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
) {
708 /* Hmm, out of range? Let's refresh the fstat() data
709 * first, before we trust that check. */
711 r
= journal_file_fstat(f
);
715 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
)
716 return -EADDRNOTAVAIL
;
719 return mmap_cache_get(f
->mmap
, f
->fd
, f
->prot
, type_to_context(type
), keep_always
, offset
, size
, &f
->last_stat
, ret
);
722 static uint64_t minimum_header_size(Object
*o
) {
724 static const uint64_t table
[] = {
725 [OBJECT_DATA
] = sizeof(DataObject
),
726 [OBJECT_FIELD
] = sizeof(FieldObject
),
727 [OBJECT_ENTRY
] = sizeof(EntryObject
),
728 [OBJECT_DATA_HASH_TABLE
] = sizeof(HashTableObject
),
729 [OBJECT_FIELD_HASH_TABLE
] = sizeof(HashTableObject
),
730 [OBJECT_ENTRY_ARRAY
] = sizeof(EntryArrayObject
),
731 [OBJECT_TAG
] = sizeof(TagObject
),
734 if (o
->object
.type
>= ELEMENTSOF(table
) || table
[o
->object
.type
] <= 0)
735 return sizeof(ObjectHeader
);
737 return table
[o
->object
.type
];
740 int journal_file_move_to_object(JournalFile
*f
, ObjectType type
, uint64_t offset
, Object
**ret
) {
749 /* Objects may only be located at multiple of 64 bit */
750 if (!VALID64(offset
))
753 /* Object may not be located in the file header */
754 if (offset
< le64toh(f
->header
->header_size
))
757 r
= journal_file_move_to(f
, type
, false, offset
, sizeof(ObjectHeader
), &t
);
762 s
= le64toh(o
->object
.size
);
764 if (s
< sizeof(ObjectHeader
))
767 if (o
->object
.type
<= OBJECT_UNUSED
)
770 if (s
< minimum_header_size(o
))
773 if (type
> OBJECT_UNUSED
&& o
->object
.type
!= type
)
776 if (s
> sizeof(ObjectHeader
)) {
777 r
= journal_file_move_to(f
, type
, false, offset
, s
, &t
);
788 static uint64_t journal_file_entry_seqnum(JournalFile
*f
, uint64_t *seqnum
) {
794 r
= le64toh(f
->header
->tail_entry_seqnum
) + 1;
797 /* If an external seqnum counter was passed, we update
798 * both the local and the external one, and set it to
799 * the maximum of both */
807 f
->header
->tail_entry_seqnum
= htole64(r
);
809 if (f
->header
->head_entry_seqnum
== 0)
810 f
->header
->head_entry_seqnum
= htole64(r
);
815 int journal_file_append_object(JournalFile
*f
, ObjectType type
, uint64_t size
, Object
**ret
, uint64_t *offset
) {
823 assert(type
> OBJECT_UNUSED
&& type
< _OBJECT_TYPE_MAX
);
824 assert(size
>= sizeof(ObjectHeader
));
828 r
= journal_file_set_online(f
);
832 p
= le64toh(f
->header
->tail_object_offset
);
834 p
= le64toh(f
->header
->header_size
);
836 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &tail
);
840 p
+= ALIGN64(le64toh(tail
->object
.size
));
843 r
= journal_file_allocate(f
, p
, size
);
847 r
= journal_file_move_to(f
, type
, false, p
, size
, &t
);
854 o
->object
.type
= type
;
855 o
->object
.size
= htole64(size
);
857 f
->header
->tail_object_offset
= htole64(p
);
858 f
->header
->n_objects
= htole64(le64toh(f
->header
->n_objects
) + 1);
866 static int journal_file_setup_data_hash_table(JournalFile
*f
) {
874 /* We estimate that we need 1 hash table entry per 768 bytes
875 of journal file and we want to make sure we never get
876 beyond 75% fill level. Calculate the hash table size for
877 the maximum file size based on these metrics. */
879 s
= (f
->metrics
.max_size
* 4 / 768 / 3) * sizeof(HashItem
);
880 if (s
< DEFAULT_DATA_HASH_TABLE_SIZE
)
881 s
= DEFAULT_DATA_HASH_TABLE_SIZE
;
883 log_debug("Reserving %"PRIu64
" entries in hash table.", s
/ sizeof(HashItem
));
885 r
= journal_file_append_object(f
,
886 OBJECT_DATA_HASH_TABLE
,
887 offsetof(Object
, hash_table
.items
) + s
,
892 memzero(o
->hash_table
.items
, s
);
894 f
->header
->data_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
895 f
->header
->data_hash_table_size
= htole64(s
);
900 static int journal_file_setup_field_hash_table(JournalFile
*f
) {
908 /* We use a fixed size hash table for the fields as this
909 * number should grow very slowly only */
911 s
= DEFAULT_FIELD_HASH_TABLE_SIZE
;
912 r
= journal_file_append_object(f
,
913 OBJECT_FIELD_HASH_TABLE
,
914 offsetof(Object
, hash_table
.items
) + s
,
919 memzero(o
->hash_table
.items
, s
);
921 f
->header
->field_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
922 f
->header
->field_hash_table_size
= htole64(s
);
927 int journal_file_map_data_hash_table(JournalFile
*f
) {
935 if (f
->data_hash_table
)
938 p
= le64toh(f
->header
->data_hash_table_offset
);
939 s
= le64toh(f
->header
->data_hash_table_size
);
941 r
= journal_file_move_to(f
,
942 OBJECT_DATA_HASH_TABLE
,
949 f
->data_hash_table
= t
;
953 int journal_file_map_field_hash_table(JournalFile
*f
) {
961 if (f
->field_hash_table
)
964 p
= le64toh(f
->header
->field_hash_table_offset
);
965 s
= le64toh(f
->header
->field_hash_table_size
);
967 r
= journal_file_move_to(f
,
968 OBJECT_FIELD_HASH_TABLE
,
975 f
->field_hash_table
= t
;
979 static int journal_file_link_field(
990 assert(f
->field_hash_table
);
994 if (o
->object
.type
!= OBJECT_FIELD
)
997 m
= le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
);
1001 /* This might alter the window we are looking at */
1002 o
->field
.next_hash_offset
= o
->field
.head_data_offset
= 0;
1005 p
= le64toh(f
->field_hash_table
[h
].tail_hash_offset
);
1007 f
->field_hash_table
[h
].head_hash_offset
= htole64(offset
);
1009 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1013 o
->field
.next_hash_offset
= htole64(offset
);
1016 f
->field_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1018 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
1019 f
->header
->n_fields
= htole64(le64toh(f
->header
->n_fields
) + 1);
1024 static int journal_file_link_data(
1035 assert(f
->data_hash_table
);
1039 if (o
->object
.type
!= OBJECT_DATA
)
1042 m
= le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
);
1046 /* This might alter the window we are looking at */
1047 o
->data
.next_hash_offset
= o
->data
.next_field_offset
= 0;
1048 o
->data
.entry_offset
= o
->data
.entry_array_offset
= 0;
1049 o
->data
.n_entries
= 0;
1052 p
= le64toh(f
->data_hash_table
[h
].tail_hash_offset
);
1054 /* Only entry in the hash table is easy */
1055 f
->data_hash_table
[h
].head_hash_offset
= htole64(offset
);
1057 /* Move back to the previous data object, to patch in
1060 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1064 o
->data
.next_hash_offset
= htole64(offset
);
1067 f
->data_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1069 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
1070 f
->header
->n_data
= htole64(le64toh(f
->header
->n_data
) + 1);
1075 int journal_file_find_field_object_with_hash(
1077 const void *field
, uint64_t size
, uint64_t hash
,
1078 Object
**ret
, uint64_t *offset
) {
1080 uint64_t p
, osize
, h
, m
;
1085 assert(field
&& size
> 0);
1087 /* If the field hash table is empty, we can't find anything */
1088 if (le64toh(f
->header
->field_hash_table_size
) <= 0)
1091 /* Map the field hash table, if it isn't mapped yet. */
1092 r
= journal_file_map_field_hash_table(f
);
1096 osize
= offsetof(Object
, field
.payload
) + size
;
1098 m
= le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
);
1103 p
= le64toh(f
->field_hash_table
[h
].head_hash_offset
);
1108 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1112 if (le64toh(o
->field
.hash
) == hash
&&
1113 le64toh(o
->object
.size
) == osize
&&
1114 memcmp(o
->field
.payload
, field
, size
) == 0) {
1124 p
= le64toh(o
->field
.next_hash_offset
);
1130 int journal_file_find_field_object(
1132 const void *field
, uint64_t size
,
1133 Object
**ret
, uint64_t *offset
) {
1138 assert(field
&& size
> 0);
1140 hash
= hash64(field
, size
);
1142 return journal_file_find_field_object_with_hash(f
,
1147 int journal_file_find_data_object_with_hash(
1149 const void *data
, uint64_t size
, uint64_t hash
,
1150 Object
**ret
, uint64_t *offset
) {
1152 uint64_t p
, osize
, h
, m
;
1157 assert(data
|| size
== 0);
1159 /* If there's no data hash table, then there's no entry. */
1160 if (le64toh(f
->header
->data_hash_table_size
) <= 0)
1163 /* Map the data hash table, if it isn't mapped yet. */
1164 r
= journal_file_map_data_hash_table(f
);
1168 osize
= offsetof(Object
, data
.payload
) + size
;
1170 m
= le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
);
1175 p
= le64toh(f
->data_hash_table
[h
].head_hash_offset
);
1180 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1184 if (le64toh(o
->data
.hash
) != hash
)
1187 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
) {
1188 #if defined(HAVE_XZ) || defined(HAVE_LZ4)
1192 l
= le64toh(o
->object
.size
);
1193 if (l
<= offsetof(Object
, data
.payload
))
1196 l
-= offsetof(Object
, data
.payload
);
1198 r
= decompress_blob(o
->object
.flags
& OBJECT_COMPRESSION_MASK
,
1199 o
->data
.payload
, l
, &f
->compress_buffer
, &f
->compress_buffer_size
, &rsize
, 0);
1203 if (rsize
== size
&&
1204 memcmp(f
->compress_buffer
, data
, size
) == 0) {
1215 return -EPROTONOSUPPORT
;
1217 } else if (le64toh(o
->object
.size
) == osize
&&
1218 memcmp(o
->data
.payload
, data
, size
) == 0) {
1230 p
= le64toh(o
->data
.next_hash_offset
);
1236 int journal_file_find_data_object(
1238 const void *data
, uint64_t size
,
1239 Object
**ret
, uint64_t *offset
) {
1244 assert(data
|| size
== 0);
1246 hash
= hash64(data
, size
);
1248 return journal_file_find_data_object_with_hash(f
,
1253 static int journal_file_append_field(
1255 const void *field
, uint64_t size
,
1256 Object
**ret
, uint64_t *offset
) {
1264 assert(field
&& size
> 0);
1266 hash
= hash64(field
, size
);
1268 r
= journal_file_find_field_object_with_hash(f
, field
, size
, hash
, &o
, &p
);
1282 osize
= offsetof(Object
, field
.payload
) + size
;
1283 r
= journal_file_append_object(f
, OBJECT_FIELD
, osize
, &o
, &p
);
1287 o
->field
.hash
= htole64(hash
);
1288 memcpy(o
->field
.payload
, field
, size
);
1290 r
= journal_file_link_field(f
, o
, p
, hash
);
1294 /* The linking might have altered the window, so let's
1295 * refresh our pointer */
1296 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1301 r
= journal_file_hmac_put_object(f
, OBJECT_FIELD
, o
, p
);
1315 static int journal_file_append_data(
1317 const void *data
, uint64_t size
,
1318 Object
**ret
, uint64_t *offset
) {
1323 int r
, compression
= 0;
1327 assert(data
|| size
== 0);
1329 hash
= hash64(data
, size
);
1331 r
= journal_file_find_data_object_with_hash(f
, data
, size
, hash
, &o
, &p
);
1345 osize
= offsetof(Object
, data
.payload
) + size
;
1346 r
= journal_file_append_object(f
, OBJECT_DATA
, osize
, &o
, &p
);
1350 o
->data
.hash
= htole64(hash
);
1352 #if defined(HAVE_XZ) || defined(HAVE_LZ4)
1353 if (JOURNAL_FILE_COMPRESS(f
) && size
>= COMPRESSION_SIZE_THRESHOLD
) {
1356 compression
= compress_blob(data
, size
, o
->data
.payload
, size
- 1, &rsize
);
1358 if (compression
>= 0) {
1359 o
->object
.size
= htole64(offsetof(Object
, data
.payload
) + rsize
);
1360 o
->object
.flags
|= compression
;
1362 log_debug("Compressed data object %"PRIu64
" -> %zu using %s",
1363 size
, rsize
, object_compressed_to_string(compression
));
1365 /* Compression didn't work, we don't really care why, let's continue without compression */
1370 if (compression
== 0)
1371 memcpy_safe(o
->data
.payload
, data
, size
);
1373 r
= journal_file_link_data(f
, o
, p
, hash
);
1378 r
= journal_file_hmac_put_object(f
, OBJECT_DATA
, o
, p
);
1383 /* The linking might have altered the window, so let's
1384 * refresh our pointer */
1385 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1392 eq
= memchr(data
, '=', size
);
1393 if (eq
&& eq
> data
) {
1397 /* Create field object ... */
1398 r
= journal_file_append_field(f
, data
, (uint8_t*) eq
- (uint8_t*) data
, &fo
, &fp
);
1402 /* ... and link it in. */
1403 o
->data
.next_field_offset
= fo
->field
.head_data_offset
;
1404 fo
->field
.head_data_offset
= le64toh(p
);
1416 uint64_t journal_file_entry_n_items(Object
*o
) {
1419 if (o
->object
.type
!= OBJECT_ENTRY
)
1422 return (le64toh(o
->object
.size
) - offsetof(Object
, entry
.items
)) / sizeof(EntryItem
);
1425 uint64_t journal_file_entry_array_n_items(Object
*o
) {
1428 if (o
->object
.type
!= OBJECT_ENTRY_ARRAY
)
1431 return (le64toh(o
->object
.size
) - offsetof(Object
, entry_array
.items
)) / sizeof(uint64_t);
1434 uint64_t journal_file_hash_table_n_items(Object
*o
) {
1437 if (o
->object
.type
!= OBJECT_DATA_HASH_TABLE
&&
1438 o
->object
.type
!= OBJECT_FIELD_HASH_TABLE
)
1441 return (le64toh(o
->object
.size
) - offsetof(Object
, hash_table
.items
)) / sizeof(HashItem
);
1444 static int link_entry_into_array(JournalFile
*f
,
1449 uint64_t n
= 0, ap
= 0, q
, i
, a
, hidx
;
1458 a
= le64toh(*first
);
1459 i
= hidx
= le64toh(*idx
);
1462 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
1466 n
= journal_file_entry_array_n_items(o
);
1468 o
->entry_array
.items
[i
] = htole64(p
);
1469 *idx
= htole64(hidx
+ 1);
1475 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
1486 r
= journal_file_append_object(f
, OBJECT_ENTRY_ARRAY
,
1487 offsetof(Object
, entry_array
.items
) + n
* sizeof(uint64_t),
1493 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY_ARRAY
, o
, q
);
1498 o
->entry_array
.items
[i
] = htole64(p
);
1501 *first
= htole64(q
);
1503 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, ap
, &o
);
1507 o
->entry_array
.next_entry_array_offset
= htole64(q
);
1510 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
1511 f
->header
->n_entry_arrays
= htole64(le64toh(f
->header
->n_entry_arrays
) + 1);
1513 *idx
= htole64(hidx
+ 1);
1518 static int link_entry_into_array_plus_one(JournalFile
*f
,
1533 *extra
= htole64(p
);
1537 i
= htole64(le64toh(*idx
) - 1);
1538 r
= link_entry_into_array(f
, first
, &i
, p
);
1543 *idx
= htole64(le64toh(*idx
) + 1);
1547 static int journal_file_link_entry_item(JournalFile
*f
, Object
*o
, uint64_t offset
, uint64_t i
) {
1554 p
= le64toh(o
->entry
.items
[i
].object_offset
);
1558 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1562 return link_entry_into_array_plus_one(f
,
1563 &o
->data
.entry_offset
,
1564 &o
->data
.entry_array_offset
,
1569 static int journal_file_link_entry(JournalFile
*f
, Object
*o
, uint64_t offset
) {
1578 if (o
->object
.type
!= OBJECT_ENTRY
)
1581 __sync_synchronize();
1583 /* Link up the entry itself */
1584 r
= link_entry_into_array(f
,
1585 &f
->header
->entry_array_offset
,
1586 &f
->header
->n_entries
,
1591 /* log_debug("=> %s seqnr=%"PRIu64" n_entries=%"PRIu64, f->path, o->entry.seqnum, f->header->n_entries); */
1593 if (f
->header
->head_entry_realtime
== 0)
1594 f
->header
->head_entry_realtime
= o
->entry
.realtime
;
1596 f
->header
->tail_entry_realtime
= o
->entry
.realtime
;
1597 f
->header
->tail_entry_monotonic
= o
->entry
.monotonic
;
1599 f
->tail_entry_monotonic_valid
= true;
1601 /* Link up the items */
1602 n
= journal_file_entry_n_items(o
);
1603 for (i
= 0; i
< n
; i
++) {
1604 r
= journal_file_link_entry_item(f
, o
, offset
, i
);
1612 static int journal_file_append_entry_internal(
1614 const dual_timestamp
*ts
,
1616 const EntryItem items
[], unsigned n_items
,
1618 Object
**ret
, uint64_t *offset
) {
1626 assert(items
|| n_items
== 0);
1629 osize
= offsetof(Object
, entry
.items
) + (n_items
* sizeof(EntryItem
));
1631 r
= journal_file_append_object(f
, OBJECT_ENTRY
, osize
, &o
, &np
);
1635 o
->entry
.seqnum
= htole64(journal_file_entry_seqnum(f
, seqnum
));
1636 memcpy_safe(o
->entry
.items
, items
, n_items
* sizeof(EntryItem
));
1637 o
->entry
.realtime
= htole64(ts
->realtime
);
1638 o
->entry
.monotonic
= htole64(ts
->monotonic
);
1639 o
->entry
.xor_hash
= htole64(xor_hash
);
1640 o
->entry
.boot_id
= f
->header
->boot_id
;
1643 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY
, o
, np
);
1648 r
= journal_file_link_entry(f
, o
, np
);
1661 void journal_file_post_change(JournalFile
*f
) {
1664 /* inotify() does not receive IN_MODIFY events from file
1665 * accesses done via mmap(). After each access we hence
1666 * trigger IN_MODIFY by truncating the journal file to its
1667 * current size which triggers IN_MODIFY. */
1669 __sync_synchronize();
1671 if (ftruncate(f
->fd
, f
->last_stat
.st_size
) < 0)
1672 log_debug_errno(errno
, "Failed to truncate file to its own size: %m");
1675 static int post_change_thunk(sd_event_source
*timer
, uint64_t usec
, void *userdata
) {
1678 journal_file_post_change(userdata
);
1683 static void schedule_post_change(JournalFile
*f
) {
1684 sd_event_source
*timer
;
1689 assert(f
->post_change_timer
);
1691 timer
= f
->post_change_timer
;
1693 r
= sd_event_source_get_enabled(timer
, &enabled
);
1695 log_debug_errno(r
, "Failed to get ftruncate timer state: %m");
1699 if (enabled
== SD_EVENT_ONESHOT
)
1702 r
= sd_event_now(sd_event_source_get_event(timer
), CLOCK_MONOTONIC
, &now
);
1704 log_debug_errno(r
, "Failed to get clock's now for scheduling ftruncate: %m");
1708 r
= sd_event_source_set_time(timer
, now
+f
->post_change_timer_period
);
1710 log_debug_errno(r
, "Failed to set time for scheduling ftruncate: %m");
1714 r
= sd_event_source_set_enabled(timer
, SD_EVENT_ONESHOT
);
1716 log_debug_errno(r
, "Failed to enable scheduled ftruncate: %m");
1723 /* On failure, let's simply post the change immediately. */
1724 journal_file_post_change(f
);
1727 /* Enable coalesced change posting in a timer on the provided sd_event instance */
1728 int journal_file_enable_post_change_timer(JournalFile
*f
, sd_event
*e
, usec_t t
) {
1729 _cleanup_(sd_event_source_unrefp
) sd_event_source
*timer
= NULL
;
1733 assert_return(!f
->post_change_timer
, -EINVAL
);
1737 r
= sd_event_add_time(e
, &timer
, CLOCK_MONOTONIC
, 0, 0, post_change_thunk
, f
);
1741 r
= sd_event_source_set_enabled(timer
, SD_EVENT_OFF
);
1745 f
->post_change_timer
= timer
;
1747 f
->post_change_timer_period
= t
;
1752 static int entry_item_cmp(const void *_a
, const void *_b
) {
1753 const EntryItem
*a
= _a
, *b
= _b
;
1755 if (le64toh(a
->object_offset
) < le64toh(b
->object_offset
))
1757 if (le64toh(a
->object_offset
) > le64toh(b
->object_offset
))
1762 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
) {
1766 uint64_t xor_hash
= 0;
1767 struct dual_timestamp _ts
;
1771 assert(iovec
|| n_iovec
== 0);
1774 dual_timestamp_get(&_ts
);
1779 r
= journal_file_maybe_append_tag(f
, ts
->realtime
);
1784 /* alloca() can't take 0, hence let's allocate at least one */
1785 items
= alloca(sizeof(EntryItem
) * MAX(1u, n_iovec
));
1787 for (i
= 0; i
< n_iovec
; i
++) {
1791 r
= journal_file_append_data(f
, iovec
[i
].iov_base
, iovec
[i
].iov_len
, &o
, &p
);
1795 xor_hash
^= le64toh(o
->data
.hash
);
1796 items
[i
].object_offset
= htole64(p
);
1797 items
[i
].hash
= o
->data
.hash
;
1800 /* Order by the position on disk, in order to improve seek
1801 * times for rotating media. */
1802 qsort_safe(items
, n_iovec
, sizeof(EntryItem
), entry_item_cmp
);
1804 r
= journal_file_append_entry_internal(f
, ts
, xor_hash
, items
, n_iovec
, seqnum
, ret
, offset
);
1806 /* If the memory mapping triggered a SIGBUS then we return an
1807 * IO error and ignore the error code passed down to us, since
1808 * it is very likely just an effect of a nullified replacement
1811 if (mmap_cache_got_sigbus(f
->mmap
, f
->fd
))
1814 if (f
->post_change_timer
)
1815 schedule_post_change(f
);
1817 journal_file_post_change(f
);
1822 typedef struct ChainCacheItem
{
1823 uint64_t first
; /* the array at the beginning of the chain */
1824 uint64_t array
; /* the cached array */
1825 uint64_t begin
; /* the first item in the cached array */
1826 uint64_t total
; /* the total number of items in all arrays before this one in the chain */
1827 uint64_t last_index
; /* the last index we looked at, to optimize locality when bisecting */
1830 static void chain_cache_put(
1837 uint64_t last_index
) {
1840 /* If the chain item to cache for this chain is the
1841 * first one it's not worth caching anything */
1845 if (ordered_hashmap_size(h
) >= CHAIN_CACHE_MAX
) {
1846 ci
= ordered_hashmap_steal_first(h
);
1849 ci
= new(ChainCacheItem
, 1);
1856 if (ordered_hashmap_put(h
, &ci
->first
, ci
) < 0) {
1861 assert(ci
->first
== first
);
1866 ci
->last_index
= last_index
;
1869 static int generic_array_get(
1873 Object
**ret
, uint64_t *offset
) {
1876 uint64_t p
= 0, a
, t
= 0;
1884 /* Try the chain cache first */
1885 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
1886 if (ci
&& i
> ci
->total
) {
1895 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
1899 k
= journal_file_entry_array_n_items(o
);
1901 p
= le64toh(o
->entry_array
.items
[i
]);
1907 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
1913 /* Let's cache this item for the next invocation */
1914 chain_cache_put(f
->chain_cache
, ci
, first
, a
, le64toh(o
->entry_array
.items
[0]), t
, i
);
1916 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
1929 static int generic_array_get_plus_one(
1934 Object
**ret
, uint64_t *offset
) {
1943 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, &o
);
1956 return generic_array_get(f
, first
, i
-1, ret
, offset
);
1965 static int generic_array_bisect(
1970 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
1971 direction_t direction
,
1976 uint64_t a
, p
, t
= 0, i
= 0, last_p
= 0, last_index
= (uint64_t) -1;
1977 bool subtract_one
= false;
1978 Object
*o
, *array
= NULL
;
1983 assert(test_object
);
1985 /* Start with the first array in the chain */
1988 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
1989 if (ci
&& n
> ci
->total
) {
1990 /* Ah, we have iterated this bisection array chain
1991 * previously! Let's see if we can skip ahead in the
1992 * chain, as far as the last time. But we can't jump
1993 * backwards in the chain, so let's check that
1996 r
= test_object(f
, ci
->begin
, needle
);
2000 if (r
== TEST_LEFT
) {
2001 /* OK, what we are looking for is right of the
2002 * begin of this EntryArray, so let's jump
2003 * straight to previously cached array in the
2009 last_index
= ci
->last_index
;
2014 uint64_t left
, right
, k
, lp
;
2016 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &array
);
2020 k
= journal_file_entry_array_n_items(array
);
2026 lp
= p
= le64toh(array
->entry_array
.items
[i
]);
2030 r
= test_object(f
, p
, needle
);
2031 if (r
== -EBADMSG
) {
2032 log_debug_errno(r
, "Encountered invalid entry while bisecting, cutting algorithm short. (1)");
2039 if (r
== TEST_FOUND
)
2040 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2042 if (r
== TEST_RIGHT
) {
2046 if (last_index
!= (uint64_t) -1) {
2047 assert(last_index
<= right
);
2049 /* If we cached the last index we
2050 * looked at, let's try to not to jump
2051 * too wildly around and see if we can
2052 * limit the range to look at early to
2053 * the immediate neighbors of the last
2054 * index we looked at. */
2056 if (last_index
> 0) {
2057 uint64_t x
= last_index
- 1;
2059 p
= le64toh(array
->entry_array
.items
[x
]);
2063 r
= test_object(f
, p
, needle
);
2067 if (r
== TEST_FOUND
)
2068 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2070 if (r
== TEST_RIGHT
)
2076 if (last_index
< right
) {
2077 uint64_t y
= last_index
+ 1;
2079 p
= le64toh(array
->entry_array
.items
[y
]);
2083 r
= test_object(f
, p
, needle
);
2087 if (r
== TEST_FOUND
)
2088 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2090 if (r
== TEST_RIGHT
)
2098 if (left
== right
) {
2099 if (direction
== DIRECTION_UP
)
2100 subtract_one
= true;
2106 assert(left
< right
);
2107 i
= (left
+ right
) / 2;
2109 p
= le64toh(array
->entry_array
.items
[i
]);
2113 r
= test_object(f
, p
, needle
);
2114 if (r
== -EBADMSG
) {
2115 log_debug_errno(r
, "Encountered invalid entry while bisecting, cutting algorithm short. (2)");
2122 if (r
== TEST_FOUND
)
2123 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2125 if (r
== TEST_RIGHT
)
2133 if (direction
== DIRECTION_UP
) {
2135 subtract_one
= true;
2146 last_index
= (uint64_t) -1;
2147 a
= le64toh(array
->entry_array
.next_entry_array_offset
);
2153 if (subtract_one
&& t
== 0 && i
== 0)
2156 /* Let's cache this item for the next invocation */
2157 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
);
2159 if (subtract_one
&& i
== 0)
2161 else if (subtract_one
)
2162 p
= le64toh(array
->entry_array
.items
[i
-1]);
2164 p
= le64toh(array
->entry_array
.items
[i
]);
2166 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2177 *idx
= t
+ i
+ (subtract_one
? -1 : 0);
2182 static int generic_array_bisect_plus_one(
2188 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
2189 direction_t direction
,
2195 bool step_back
= false;
2199 assert(test_object
);
2204 /* This bisects the array in object 'first', but first checks
2206 r
= test_object(f
, extra
, needle
);
2210 if (r
== TEST_FOUND
)
2211 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2213 /* if we are looking with DIRECTION_UP then we need to first
2214 see if in the actual array there is a matching entry, and
2215 return the last one of that. But if there isn't any we need
2216 to return this one. Hence remember this, and return it
2219 step_back
= direction
== DIRECTION_UP
;
2221 if (r
== TEST_RIGHT
) {
2222 if (direction
== DIRECTION_DOWN
)
2228 r
= generic_array_bisect(f
, first
, n
-1, needle
, test_object
, direction
, ret
, offset
, idx
);
2230 if (r
== 0 && step_back
)
2239 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, &o
);
2255 _pure_
static int test_object_offset(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2261 else if (p
< needle
)
2267 static int test_object_seqnum(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2274 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2278 if (le64toh(o
->entry
.seqnum
) == needle
)
2280 else if (le64toh(o
->entry
.seqnum
) < needle
)
2286 int journal_file_move_to_entry_by_seqnum(
2289 direction_t direction
,
2295 return generic_array_bisect(f
,
2296 le64toh(f
->header
->entry_array_offset
),
2297 le64toh(f
->header
->n_entries
),
2304 static int test_object_realtime(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2311 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2315 if (le64toh(o
->entry
.realtime
) == needle
)
2317 else if (le64toh(o
->entry
.realtime
) < needle
)
2323 int journal_file_move_to_entry_by_realtime(
2326 direction_t direction
,
2332 return generic_array_bisect(f
,
2333 le64toh(f
->header
->entry_array_offset
),
2334 le64toh(f
->header
->n_entries
),
2336 test_object_realtime
,
2341 static int test_object_monotonic(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2348 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2352 if (le64toh(o
->entry
.monotonic
) == needle
)
2354 else if (le64toh(o
->entry
.monotonic
) < needle
)
2360 static int find_data_object_by_boot_id(
2366 char t
[sizeof("_BOOT_ID=")-1 + 32 + 1] = "_BOOT_ID=";
2368 sd_id128_to_string(boot_id
, t
+ 9);
2369 return journal_file_find_data_object(f
, t
, sizeof(t
) - 1, o
, b
);
2372 int journal_file_move_to_entry_by_monotonic(
2376 direction_t direction
,
2385 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, NULL
);
2391 return generic_array_bisect_plus_one(f
,
2392 le64toh(o
->data
.entry_offset
),
2393 le64toh(o
->data
.entry_array_offset
),
2394 le64toh(o
->data
.n_entries
),
2396 test_object_monotonic
,
2401 void journal_file_reset_location(JournalFile
*f
) {
2402 f
->location_type
= LOCATION_HEAD
;
2403 f
->current_offset
= 0;
2404 f
->current_seqnum
= 0;
2405 f
->current_realtime
= 0;
2406 f
->current_monotonic
= 0;
2407 zero(f
->current_boot_id
);
2408 f
->current_xor_hash
= 0;
2411 void journal_file_save_location(JournalFile
*f
, Object
*o
, uint64_t offset
) {
2412 f
->location_type
= LOCATION_SEEK
;
2413 f
->current_offset
= offset
;
2414 f
->current_seqnum
= le64toh(o
->entry
.seqnum
);
2415 f
->current_realtime
= le64toh(o
->entry
.realtime
);
2416 f
->current_monotonic
= le64toh(o
->entry
.monotonic
);
2417 f
->current_boot_id
= o
->entry
.boot_id
;
2418 f
->current_xor_hash
= le64toh(o
->entry
.xor_hash
);
2421 int journal_file_compare_locations(JournalFile
*af
, JournalFile
*bf
) {
2426 assert(af
->location_type
== LOCATION_SEEK
);
2427 assert(bf
->location_type
== LOCATION_SEEK
);
2429 /* If contents and timestamps match, these entries are
2430 * identical, even if the seqnum does not match */
2431 if (sd_id128_equal(af
->current_boot_id
, bf
->current_boot_id
) &&
2432 af
->current_monotonic
== bf
->current_monotonic
&&
2433 af
->current_realtime
== bf
->current_realtime
&&
2434 af
->current_xor_hash
== bf
->current_xor_hash
)
2437 if (sd_id128_equal(af
->header
->seqnum_id
, bf
->header
->seqnum_id
)) {
2439 /* If this is from the same seqnum source, compare
2441 if (af
->current_seqnum
< bf
->current_seqnum
)
2443 if (af
->current_seqnum
> bf
->current_seqnum
)
2446 /* Wow! This is weird, different data but the same
2447 * seqnums? Something is borked, but let's make the
2448 * best of it and compare by time. */
2451 if (sd_id128_equal(af
->current_boot_id
, bf
->current_boot_id
)) {
2453 /* If the boot id matches, compare monotonic time */
2454 if (af
->current_monotonic
< bf
->current_monotonic
)
2456 if (af
->current_monotonic
> bf
->current_monotonic
)
2460 /* Otherwise, compare UTC time */
2461 if (af
->current_realtime
< bf
->current_realtime
)
2463 if (af
->current_realtime
> bf
->current_realtime
)
2466 /* Finally, compare by contents */
2467 if (af
->current_xor_hash
< bf
->current_xor_hash
)
2469 if (af
->current_xor_hash
> bf
->current_xor_hash
)
2475 int journal_file_next_entry(
2478 direction_t direction
,
2479 Object
**ret
, uint64_t *offset
) {
2487 n
= le64toh(f
->header
->n_entries
);
2492 i
= direction
== DIRECTION_DOWN
? 0 : n
- 1;
2494 r
= generic_array_bisect(f
,
2495 le64toh(f
->header
->entry_array_offset
),
2496 le64toh(f
->header
->n_entries
),
2505 if (direction
== DIRECTION_DOWN
) {
2518 /* And jump to it */
2519 r
= generic_array_get(f
,
2520 le64toh(f
->header
->entry_array_offset
),
2523 if (r
== -EBADMSG
&& direction
== DIRECTION_DOWN
) {
2524 /* Special case: when we iterate throught the journal file linearly, and hit an entry we can't read,
2525 * consider this the end of the journal file. */
2526 log_debug_errno(r
, "Encountered entry we can't read while iterating through journal file. Considering this the end of the file.");
2533 (direction
== DIRECTION_DOWN
? ofs
<= p
: ofs
>= p
)) {
2534 log_debug("%s: entry array corrupted at entry %" PRIu64
, f
->path
, i
);
2544 int journal_file_next_entry_for_data(
2546 Object
*o
, uint64_t p
,
2547 uint64_t data_offset
,
2548 direction_t direction
,
2549 Object
**ret
, uint64_t *offset
) {
2556 assert(p
> 0 || !o
);
2558 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2562 n
= le64toh(d
->data
.n_entries
);
2567 i
= direction
== DIRECTION_DOWN
? 0 : n
- 1;
2569 if (o
->object
.type
!= OBJECT_ENTRY
)
2572 r
= generic_array_bisect_plus_one(f
,
2573 le64toh(d
->data
.entry_offset
),
2574 le64toh(d
->data
.entry_array_offset
),
2575 le64toh(d
->data
.n_entries
),
2585 if (direction
== DIRECTION_DOWN
) {
2599 return generic_array_get_plus_one(f
,
2600 le64toh(d
->data
.entry_offset
),
2601 le64toh(d
->data
.entry_array_offset
),
2606 int journal_file_move_to_entry_by_offset_for_data(
2608 uint64_t data_offset
,
2610 direction_t direction
,
2611 Object
**ret
, uint64_t *offset
) {
2618 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2622 return generic_array_bisect_plus_one(f
,
2623 le64toh(d
->data
.entry_offset
),
2624 le64toh(d
->data
.entry_array_offset
),
2625 le64toh(d
->data
.n_entries
),
2632 int journal_file_move_to_entry_by_monotonic_for_data(
2634 uint64_t data_offset
,
2637 direction_t direction
,
2638 Object
**ret
, uint64_t *offset
) {
2646 /* First, seek by time */
2647 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, &b
);
2653 r
= generic_array_bisect_plus_one(f
,
2654 le64toh(o
->data
.entry_offset
),
2655 le64toh(o
->data
.entry_array_offset
),
2656 le64toh(o
->data
.n_entries
),
2658 test_object_monotonic
,
2664 /* And now, continue seeking until we find an entry that
2665 * exists in both bisection arrays */
2671 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2675 r
= generic_array_bisect_plus_one(f
,
2676 le64toh(d
->data
.entry_offset
),
2677 le64toh(d
->data
.entry_array_offset
),
2678 le64toh(d
->data
.n_entries
),
2686 r
= journal_file_move_to_object(f
, OBJECT_DATA
, b
, &o
);
2690 r
= generic_array_bisect_plus_one(f
,
2691 le64toh(o
->data
.entry_offset
),
2692 le64toh(o
->data
.entry_array_offset
),
2693 le64toh(o
->data
.n_entries
),
2715 int journal_file_move_to_entry_by_seqnum_for_data(
2717 uint64_t data_offset
,
2719 direction_t direction
,
2720 Object
**ret
, uint64_t *offset
) {
2727 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2731 return generic_array_bisect_plus_one(f
,
2732 le64toh(d
->data
.entry_offset
),
2733 le64toh(d
->data
.entry_array_offset
),
2734 le64toh(d
->data
.n_entries
),
2741 int journal_file_move_to_entry_by_realtime_for_data(
2743 uint64_t data_offset
,
2745 direction_t direction
,
2746 Object
**ret
, uint64_t *offset
) {
2753 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2757 return generic_array_bisect_plus_one(f
,
2758 le64toh(d
->data
.entry_offset
),
2759 le64toh(d
->data
.entry_array_offset
),
2760 le64toh(d
->data
.n_entries
),
2762 test_object_realtime
,
2767 void journal_file_dump(JournalFile
*f
) {
2775 journal_file_print_header(f
);
2777 p
= le64toh(f
->header
->header_size
);
2779 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &o
);
2783 switch (o
->object
.type
) {
2786 printf("Type: OBJECT_UNUSED\n");
2790 printf("Type: OBJECT_DATA\n");
2794 printf("Type: OBJECT_FIELD\n");
2798 printf("Type: OBJECT_ENTRY seqnum=%"PRIu64
" monotonic=%"PRIu64
" realtime=%"PRIu64
"\n",
2799 le64toh(o
->entry
.seqnum
),
2800 le64toh(o
->entry
.monotonic
),
2801 le64toh(o
->entry
.realtime
));
2804 case OBJECT_FIELD_HASH_TABLE
:
2805 printf("Type: OBJECT_FIELD_HASH_TABLE\n");
2808 case OBJECT_DATA_HASH_TABLE
:
2809 printf("Type: OBJECT_DATA_HASH_TABLE\n");
2812 case OBJECT_ENTRY_ARRAY
:
2813 printf("Type: OBJECT_ENTRY_ARRAY\n");
2817 printf("Type: OBJECT_TAG seqnum=%"PRIu64
" epoch=%"PRIu64
"\n",
2818 le64toh(o
->tag
.seqnum
),
2819 le64toh(o
->tag
.epoch
));
2823 printf("Type: unknown (%i)\n", o
->object
.type
);
2827 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
)
2828 printf("Flags: %s\n",
2829 object_compressed_to_string(o
->object
.flags
& OBJECT_COMPRESSION_MASK
));
2831 if (p
== le64toh(f
->header
->tail_object_offset
))
2834 p
= p
+ ALIGN64(le64toh(o
->object
.size
));
2839 log_error("File corrupt");
2842 static const char* format_timestamp_safe(char *buf
, size_t l
, usec_t t
) {
2845 x
= format_timestamp(buf
, l
, t
);
2851 void journal_file_print_header(JournalFile
*f
) {
2852 char a
[33], b
[33], c
[33], d
[33];
2853 char x
[FORMAT_TIMESTAMP_MAX
], y
[FORMAT_TIMESTAMP_MAX
], z
[FORMAT_TIMESTAMP_MAX
];
2855 char bytes
[FORMAT_BYTES_MAX
];
2860 printf("File Path: %s\n"
2864 "Sequential Number ID: %s\n"
2866 "Compatible Flags:%s%s\n"
2867 "Incompatible Flags:%s%s%s\n"
2868 "Header size: %"PRIu64
"\n"
2869 "Arena size: %"PRIu64
"\n"
2870 "Data Hash Table Size: %"PRIu64
"\n"
2871 "Field Hash Table Size: %"PRIu64
"\n"
2872 "Rotate Suggested: %s\n"
2873 "Head Sequential Number: %"PRIu64
" (%"PRIx64
")\n"
2874 "Tail Sequential Number: %"PRIu64
" (%"PRIx64
")\n"
2875 "Head Realtime Timestamp: %s (%"PRIx64
")\n"
2876 "Tail Realtime Timestamp: %s (%"PRIx64
")\n"
2877 "Tail Monotonic Timestamp: %s (%"PRIx64
")\n"
2878 "Objects: %"PRIu64
"\n"
2879 "Entry Objects: %"PRIu64
"\n",
2881 sd_id128_to_string(f
->header
->file_id
, a
),
2882 sd_id128_to_string(f
->header
->machine_id
, b
),
2883 sd_id128_to_string(f
->header
->boot_id
, c
),
2884 sd_id128_to_string(f
->header
->seqnum_id
, d
),
2885 f
->header
->state
== STATE_OFFLINE
? "OFFLINE" :
2886 f
->header
->state
== STATE_ONLINE
? "ONLINE" :
2887 f
->header
->state
== STATE_ARCHIVED
? "ARCHIVED" : "UNKNOWN",
2888 JOURNAL_HEADER_SEALED(f
->header
) ? " SEALED" : "",
2889 (le32toh(f
->header
->compatible_flags
) & ~HEADER_COMPATIBLE_ANY
) ? " ???" : "",
2890 JOURNAL_HEADER_COMPRESSED_XZ(f
->header
) ? " COMPRESSED-XZ" : "",
2891 JOURNAL_HEADER_COMPRESSED_LZ4(f
->header
) ? " COMPRESSED-LZ4" : "",
2892 (le32toh(f
->header
->incompatible_flags
) & ~HEADER_INCOMPATIBLE_ANY
) ? " ???" : "",
2893 le64toh(f
->header
->header_size
),
2894 le64toh(f
->header
->arena_size
),
2895 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
2896 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
),
2897 yes_no(journal_file_rotate_suggested(f
, 0)),
2898 le64toh(f
->header
->head_entry_seqnum
), le64toh(f
->header
->head_entry_seqnum
),
2899 le64toh(f
->header
->tail_entry_seqnum
), le64toh(f
->header
->tail_entry_seqnum
),
2900 format_timestamp_safe(x
, sizeof(x
), le64toh(f
->header
->head_entry_realtime
)), le64toh(f
->header
->head_entry_realtime
),
2901 format_timestamp_safe(y
, sizeof(y
), le64toh(f
->header
->tail_entry_realtime
)), le64toh(f
->header
->tail_entry_realtime
),
2902 format_timespan(z
, sizeof(z
), le64toh(f
->header
->tail_entry_monotonic
), USEC_PER_MSEC
), le64toh(f
->header
->tail_entry_monotonic
),
2903 le64toh(f
->header
->n_objects
),
2904 le64toh(f
->header
->n_entries
));
2906 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
2907 printf("Data Objects: %"PRIu64
"\n"
2908 "Data Hash Table Fill: %.1f%%\n",
2909 le64toh(f
->header
->n_data
),
2910 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))));
2912 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
2913 printf("Field Objects: %"PRIu64
"\n"
2914 "Field Hash Table Fill: %.1f%%\n",
2915 le64toh(f
->header
->n_fields
),
2916 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))));
2918 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_tags
))
2919 printf("Tag Objects: %"PRIu64
"\n",
2920 le64toh(f
->header
->n_tags
));
2921 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
2922 printf("Entry Array Objects: %"PRIu64
"\n",
2923 le64toh(f
->header
->n_entry_arrays
));
2925 if (fstat(f
->fd
, &st
) >= 0)
2926 printf("Disk usage: %s\n", format_bytes(bytes
, sizeof(bytes
), (uint64_t) st
.st_blocks
* 512ULL));
2929 static int journal_file_warn_btrfs(JournalFile
*f
) {
2935 /* Before we write anything, check if the COW logic is turned
2936 * off on btrfs. Given our write pattern that is quite
2937 * unfriendly to COW file systems this should greatly improve
2938 * performance on COW file systems, such as btrfs, at the
2939 * expense of data integrity features (which shouldn't be too
2940 * bad, given that we do our own checksumming). */
2942 r
= btrfs_is_filesystem(f
->fd
);
2944 return log_warning_errno(r
, "Failed to determine if journal is on btrfs: %m");
2948 r
= read_attr_fd(f
->fd
, &attrs
);
2950 return log_warning_errno(r
, "Failed to read file attributes: %m");
2952 if (attrs
& FS_NOCOW_FL
) {
2953 log_debug("Detected btrfs file system with copy-on-write disabled, all is good.");
2957 log_notice("Creating journal file %s on a btrfs file system, and copy-on-write is enabled. "
2958 "This is likely to slow down journal access substantially, please consider turning "
2959 "off the copy-on-write file attribute on the journal directory, using chattr +C.", f
->path
);
2964 int journal_file_open(
2971 JournalMetrics
*metrics
,
2972 MMapCache
*mmap_cache
,
2973 Set
*deferred_closes
,
2974 JournalFile
*template,
2975 JournalFile
**ret
) {
2977 bool newly_created
= false;
2983 assert(fd
>= 0 || fname
);
2985 if ((flags
& O_ACCMODE
) != O_RDONLY
&&
2986 (flags
& O_ACCMODE
) != O_RDWR
)
2990 if (!endswith(fname
, ".journal") &&
2991 !endswith(fname
, ".journal~"))
2995 f
= new0(JournalFile
, 1);
3003 f
->prot
= prot_from_flags(flags
);
3004 f
->writable
= (flags
& O_ACCMODE
) != O_RDONLY
;
3005 #if defined(HAVE_LZ4)
3006 f
->compress_lz4
= compress
;
3007 #elif defined(HAVE_XZ)
3008 f
->compress_xz
= compress
;
3015 f
->mmap
= mmap_cache_ref(mmap_cache
);
3017 f
->mmap
= mmap_cache_new();
3025 f
->path
= strdup(fname
);
3026 else /* If we don't know the path, fill in something explanatory and vaguely useful */
3027 asprintf(&f
->path
, "/proc/self/%i", fd
);
3033 f
->chain_cache
= ordered_hashmap_new(&uint64_hash_ops
);
3034 if (!f
->chain_cache
) {
3040 f
->fd
= open(f
->path
, f
->flags
|O_CLOEXEC
, f
->mode
);
3046 /* fds we opened here by us should also be closed by us. */
3050 r
= journal_file_fstat(f
);
3054 if (f
->last_stat
.st_size
== 0 && f
->writable
) {
3056 (void) journal_file_warn_btrfs(f
);
3058 /* Let's attach the creation time to the journal file,
3059 * so that the vacuuming code knows the age of this
3060 * file even if the file might end up corrupted one
3061 * day... Ideally we'd just use the creation time many
3062 * file systems maintain for each file, but there is
3063 * currently no usable API to query this, hence let's
3064 * emulate this via extended attributes. If extended
3065 * attributes are not supported we'll just skip this,
3066 * and rely solely on mtime/atime/ctime of the file. */
3068 fd_setcrtime(f
->fd
, 0);
3071 /* Try to load the FSPRG state, and if we can't, then
3072 * just don't do sealing */
3074 r
= journal_file_fss_load(f
);
3080 r
= journal_file_init_header(f
, template);
3084 r
= journal_file_fstat(f
);
3088 newly_created
= true;
3091 if (f
->last_stat
.st_size
< (off_t
) HEADER_SIZE_MIN
) {
3096 r
= mmap_cache_get(f
->mmap
, f
->fd
, f
->prot
, CONTEXT_HEADER
, true, 0, PAGE_ALIGN(sizeof(Header
)), &f
->last_stat
, &h
);
3102 if (!newly_created
) {
3103 if (deferred_closes
)
3104 journal_file_close_set(deferred_closes
);
3106 r
= journal_file_verify_header(f
);
3112 if (!newly_created
&& f
->writable
) {
3113 r
= journal_file_fss_load(f
);
3121 journal_default_metrics(metrics
, f
->fd
);
3122 f
->metrics
= *metrics
;
3123 } else if (template)
3124 f
->metrics
= template->metrics
;
3126 r
= journal_file_refresh_header(f
);
3132 r
= journal_file_hmac_setup(f
);
3137 if (newly_created
) {
3138 r
= journal_file_setup_field_hash_table(f
);
3142 r
= journal_file_setup_data_hash_table(f
);
3147 r
= journal_file_append_first_tag(f
);
3153 if (mmap_cache_got_sigbus(f
->mmap
, f
->fd
)) {
3158 if (template && template->post_change_timer
) {
3159 r
= journal_file_enable_post_change_timer(
3161 sd_event_source_get_event(template->post_change_timer
),
3162 template->post_change_timer_period
);
3168 /* The file is opened now successfully, thus we take possession of any passed in fd. */
3175 if (f
->fd
>= 0 && mmap_cache_got_sigbus(f
->mmap
, f
->fd
))
3178 (void) journal_file_close(f
);
3183 int journal_file_rotate(JournalFile
**f
, bool compress
, bool seal
, Set
*deferred_closes
) {
3184 _cleanup_free_
char *p
= NULL
;
3186 JournalFile
*old_file
, *new_file
= NULL
;
3194 if (!old_file
->writable
)
3197 /* Is this a journal file that was passed to us as fd? If so, we synthesized a path name for it, and we refuse
3198 * rotation, since we don't know the actual path, and couldn't rename the file hence.*/
3199 if (path_startswith(old_file
->path
, "/proc/self/fd"))
3202 if (!endswith(old_file
->path
, ".journal"))
3205 l
= strlen(old_file
->path
);
3206 r
= asprintf(&p
, "%.*s@" SD_ID128_FORMAT_STR
"-%016"PRIx64
"-%016"PRIx64
".journal",
3207 (int) l
- 8, old_file
->path
,
3208 SD_ID128_FORMAT_VAL(old_file
->header
->seqnum_id
),
3209 le64toh((*f
)->header
->head_entry_seqnum
),
3210 le64toh((*f
)->header
->head_entry_realtime
));
3214 /* Try to rename the file to the archived version. If the file
3215 * already was deleted, we'll get ENOENT, let's ignore that
3217 r
= rename(old_file
->path
, p
);
3218 if (r
< 0 && errno
!= ENOENT
)
3221 /* Sync the rename to disk */
3222 (void) fsync_directory_of_file(old_file
->fd
);
3224 /* Set as archive so offlining commits w/state=STATE_ARCHIVED.
3225 * Previously we would set old_file->header->state to STATE_ARCHIVED directly here,
3226 * but journal_file_set_offline() short-circuits when state != STATE_ONLINE, which
3227 * would result in the rotated journal never getting fsync() called before closing.
3228 * Now we simply queue the archive state by setting an archive bit, leaving the state
3229 * as STATE_ONLINE so proper offlining occurs. */
3230 old_file
->archive
= true;
3232 /* Currently, btrfs is not very good with out write patterns
3233 * and fragments heavily. Let's defrag our journal files when
3234 * we archive them */
3235 old_file
->defrag_on_close
= true;
3237 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
);
3239 if (deferred_closes
&&
3240 set_put(deferred_closes
, old_file
) >= 0)
3241 (void) journal_file_set_offline(old_file
, false);
3243 (void) journal_file_close(old_file
);
3249 int journal_file_open_reliably(
3255 JournalMetrics
*metrics
,
3256 MMapCache
*mmap_cache
,
3257 Set
*deferred_closes
,
3258 JournalFile
*template,
3259 JournalFile
**ret
) {
3263 _cleanup_free_
char *p
= NULL
;
3265 r
= journal_file_open(-1, fname
, flags
, mode
, compress
, seal
, metrics
, mmap_cache
, deferred_closes
, template, ret
);
3267 -EBADMSG
, /* corrupted */
3268 -ENODATA
, /* truncated */
3269 -EHOSTDOWN
, /* other machine */
3270 -EPROTONOSUPPORT
, /* incompatible feature */
3271 -EBUSY
, /* unclean shutdown */
3272 -ESHUTDOWN
, /* already archived */
3273 -EIO
, /* IO error, including SIGBUS on mmap */
3274 -EIDRM
/* File has been deleted */))
3277 if ((flags
& O_ACCMODE
) == O_RDONLY
)
3280 if (!(flags
& O_CREAT
))
3283 if (!endswith(fname
, ".journal"))
3286 /* The file is corrupted. Rotate it away and try it again (but only once) */
3289 if (asprintf(&p
, "%.*s@%016"PRIx64
"-%016"PRIx64
".journal~",
3291 now(CLOCK_REALTIME
),
3295 if (rename(fname
, p
) < 0)
3298 /* btrfs doesn't cope well with our write pattern and
3299 * fragments heavily. Let's defrag all files we rotate */
3301 (void) chattr_path(p
, 0, FS_NOCOW_FL
);
3302 (void) btrfs_defrag(p
);
3304 log_warning_errno(r
, "File %s corrupted or uncleanly shut down, renaming and replacing.", fname
);
3306 return journal_file_open(-1, fname
, flags
, mode
, compress
, seal
, metrics
, mmap_cache
, deferred_closes
, template, ret
);
3309 int journal_file_copy_entry(JournalFile
*from
, JournalFile
*to
, Object
*o
, uint64_t p
, uint64_t *seqnum
, Object
**ret
, uint64_t *offset
) {
3311 uint64_t q
, xor_hash
= 0;
3324 ts
.monotonic
= le64toh(o
->entry
.monotonic
);
3325 ts
.realtime
= le64toh(o
->entry
.realtime
);
3327 n
= journal_file_entry_n_items(o
);
3328 /* alloca() can't take 0, hence let's allocate at least one */
3329 items
= alloca(sizeof(EntryItem
) * MAX(1u, n
));
3331 for (i
= 0; i
< n
; i
++) {
3338 q
= le64toh(o
->entry
.items
[i
].object_offset
);
3339 le_hash
= o
->entry
.items
[i
].hash
;
3341 r
= journal_file_move_to_object(from
, OBJECT_DATA
, q
, &o
);
3345 if (le_hash
!= o
->data
.hash
)
3348 l
= le64toh(o
->object
.size
) - offsetof(Object
, data
.payload
);
3351 /* We hit the limit on 32bit machines */
3352 if ((uint64_t) t
!= l
)
3355 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
) {
3356 #if defined(HAVE_XZ) || defined(HAVE_LZ4)
3359 r
= decompress_blob(o
->object
.flags
& OBJECT_COMPRESSION_MASK
,
3360 o
->data
.payload
, l
, &from
->compress_buffer
, &from
->compress_buffer_size
, &rsize
, 0);
3364 data
= from
->compress_buffer
;
3367 return -EPROTONOSUPPORT
;
3370 data
= o
->data
.payload
;
3372 r
= journal_file_append_data(to
, data
, l
, &u
, &h
);
3376 xor_hash
^= le64toh(u
->data
.hash
);
3377 items
[i
].object_offset
= htole64(h
);
3378 items
[i
].hash
= u
->data
.hash
;
3380 r
= journal_file_move_to_object(from
, OBJECT_ENTRY
, p
, &o
);
3385 r
= journal_file_append_entry_internal(to
, &ts
, xor_hash
, items
, n
, seqnum
, ret
, offset
);
3387 if (mmap_cache_got_sigbus(to
->mmap
, to
->fd
))
3393 void journal_reset_metrics(JournalMetrics
*m
) {
3396 /* Set everything to "pick automatic values". */
3398 *m
= (JournalMetrics
) {
3399 .min_use
= (uint64_t) -1,
3400 .max_use
= (uint64_t) -1,
3401 .min_size
= (uint64_t) -1,
3402 .max_size
= (uint64_t) -1,
3403 .keep_free
= (uint64_t) -1,
3404 .n_max_files
= (uint64_t) -1,
3408 void journal_default_metrics(JournalMetrics
*m
, int fd
) {
3409 char a
[FORMAT_BYTES_MAX
], b
[FORMAT_BYTES_MAX
], c
[FORMAT_BYTES_MAX
], d
[FORMAT_BYTES_MAX
], e
[FORMAT_BYTES_MAX
];
3416 if (fstatvfs(fd
, &ss
) >= 0)
3417 fs_size
= ss
.f_frsize
* ss
.f_blocks
;
3419 log_debug_errno(errno
, "Failed to detremine disk size: %m");
3423 if (m
->max_use
== (uint64_t) -1) {
3426 m
->max_use
= PAGE_ALIGN(fs_size
/ 10); /* 10% of file system size */
3428 if (m
->max_use
> DEFAULT_MAX_USE_UPPER
)
3429 m
->max_use
= DEFAULT_MAX_USE_UPPER
;
3431 if (m
->max_use
< DEFAULT_MAX_USE_LOWER
)
3432 m
->max_use
= DEFAULT_MAX_USE_LOWER
;
3434 m
->max_use
= DEFAULT_MAX_USE_LOWER
;
3436 m
->max_use
= PAGE_ALIGN(m
->max_use
);
3438 if (m
->max_use
!= 0 && m
->max_use
< JOURNAL_FILE_SIZE_MIN
*2)
3439 m
->max_use
= JOURNAL_FILE_SIZE_MIN
*2;
3442 if (m
->min_use
== (uint64_t) -1)
3443 m
->min_use
= DEFAULT_MIN_USE
;
3445 if (m
->min_use
> m
->max_use
)
3446 m
->min_use
= m
->max_use
;
3448 if (m
->max_size
== (uint64_t) -1) {
3449 m
->max_size
= PAGE_ALIGN(m
->max_use
/ 8); /* 8 chunks */
3451 if (m
->max_size
> DEFAULT_MAX_SIZE_UPPER
)
3452 m
->max_size
= DEFAULT_MAX_SIZE_UPPER
;
3454 m
->max_size
= PAGE_ALIGN(m
->max_size
);
3456 if (m
->max_size
!= 0) {
3457 if (m
->max_size
< JOURNAL_FILE_SIZE_MIN
)
3458 m
->max_size
= JOURNAL_FILE_SIZE_MIN
;
3460 if (m
->max_use
!= 0 && m
->max_size
*2 > m
->max_use
)
3461 m
->max_use
= m
->max_size
*2;
3464 if (m
->min_size
== (uint64_t) -1)
3465 m
->min_size
= JOURNAL_FILE_SIZE_MIN
;
3467 m
->min_size
= PAGE_ALIGN(m
->min_size
);
3469 if (m
->min_size
< JOURNAL_FILE_SIZE_MIN
)
3470 m
->min_size
= JOURNAL_FILE_SIZE_MIN
;
3472 if (m
->max_size
!= 0 && m
->min_size
> m
->max_size
)
3473 m
->max_size
= m
->min_size
;
3476 if (m
->keep_free
== (uint64_t) -1) {
3479 m
->keep_free
= PAGE_ALIGN(fs_size
* 3 / 20); /* 15% of file system size */
3481 if (m
->keep_free
> DEFAULT_KEEP_FREE_UPPER
)
3482 m
->keep_free
= DEFAULT_KEEP_FREE_UPPER
;
3485 m
->keep_free
= DEFAULT_KEEP_FREE
;
3488 if (m
->n_max_files
== (uint64_t) -1)
3489 m
->n_max_files
= DEFAULT_N_MAX_FILES
;
3491 log_debug("Fixed min_use=%s max_use=%s max_size=%s min_size=%s keep_free=%s n_max_files=%" PRIu64
,
3492 format_bytes(a
, sizeof(a
), m
->min_use
),
3493 format_bytes(b
, sizeof(b
), m
->max_use
),
3494 format_bytes(c
, sizeof(c
), m
->max_size
),
3495 format_bytes(d
, sizeof(d
), m
->min_size
),
3496 format_bytes(e
, sizeof(e
), m
->keep_free
),
3500 int journal_file_get_cutoff_realtime_usec(JournalFile
*f
, usec_t
*from
, usec_t
*to
) {
3506 if (f
->header
->head_entry_realtime
== 0)
3509 *from
= le64toh(f
->header
->head_entry_realtime
);
3513 if (f
->header
->tail_entry_realtime
== 0)
3516 *to
= le64toh(f
->header
->tail_entry_realtime
);
3522 int journal_file_get_cutoff_monotonic_usec(JournalFile
*f
, sd_id128_t boot_id
, usec_t
*from
, usec_t
*to
) {
3530 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, &p
);
3534 if (le64toh(o
->data
.n_entries
) <= 0)
3538 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, le64toh(o
->data
.entry_offset
), &o
);
3542 *from
= le64toh(o
->entry
.monotonic
);
3546 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
3550 r
= generic_array_get_plus_one(f
,
3551 le64toh(o
->data
.entry_offset
),
3552 le64toh(o
->data
.entry_array_offset
),
3553 le64toh(o
->data
.n_entries
)-1,
3558 *to
= le64toh(o
->entry
.monotonic
);
3564 bool journal_file_rotate_suggested(JournalFile
*f
, usec_t max_file_usec
) {
3568 /* If we gained new header fields we gained new features,
3569 * hence suggest a rotation */
3570 if (le64toh(f
->header
->header_size
) < sizeof(Header
)) {
3571 log_debug("%s uses an outdated header, suggesting rotation.", f
->path
);
3575 /* Let's check if the hash tables grew over a certain fill
3576 * level (75%, borrowing this value from Java's hash table
3577 * implementation), and if so suggest a rotation. To calculate
3578 * the fill level we need the n_data field, which only exists
3579 * in newer versions. */
3581 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
3582 if (le64toh(f
->header
->n_data
) * 4ULL > (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
3583 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.",
3585 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))),
3586 le64toh(f
->header
->n_data
),
3587 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
3588 (unsigned long long) f
->last_stat
.st_size
,
3589 f
->last_stat
.st_size
/ le64toh(f
->header
->n_data
));
3593 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
3594 if (le64toh(f
->header
->n_fields
) * 4ULL > (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
3595 log_debug("Field hash table of %s has a fill level at %.1f (%"PRIu64
" of %"PRIu64
" items), suggesting rotation.",
3597 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))),
3598 le64toh(f
->header
->n_fields
),
3599 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
));
3603 /* Are the data objects properly indexed by field objects? */
3604 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
) &&
3605 JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
) &&
3606 le64toh(f
->header
->n_data
) > 0 &&
3607 le64toh(f
->header
->n_fields
) == 0)
3610 if (max_file_usec
> 0) {
3613 h
= le64toh(f
->header
->head_entry_realtime
);
3614 t
= now(CLOCK_REALTIME
);
3616 if (h
> 0 && t
> h
+ max_file_usec
)