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
)))
571 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
);
594 /* Don't permit appending to files from the future. Because otherwise the realtime timestamps wouldn't
595 * be strictly ordered in the entries in the file anymore, and we can't have that since it breaks
597 if (le64toh(f
->header
->tail_entry_realtime
) > now(CLOCK_REALTIME
)) {
598 log_debug("Journal file %s is from the future, refusing to append new data to it that'd be older.", f
->path
);
603 f
->compress_xz
= JOURNAL_HEADER_COMPRESSED_XZ(f
->header
);
604 f
->compress_lz4
= JOURNAL_HEADER_COMPRESSED_LZ4(f
->header
);
606 f
->seal
= JOURNAL_HEADER_SEALED(f
->header
);
611 static int journal_file_fstat(JournalFile
*f
) {
615 if (fstat(f
->fd
, &f
->last_stat
) < 0)
618 f
->last_stat_usec
= now(CLOCK_MONOTONIC
);
620 /* Refuse appending to files that are already deleted */
621 if (f
->last_stat
.st_nlink
<= 0)
627 static int journal_file_allocate(JournalFile
*f
, uint64_t offset
, uint64_t size
) {
628 uint64_t old_size
, new_size
;
634 /* We assume that this file is not sparse, and we know that
635 * for sure, since we always call posix_fallocate()
638 if (mmap_cache_got_sigbus(f
->mmap
, f
->fd
))
642 le64toh(f
->header
->header_size
) +
643 le64toh(f
->header
->arena_size
);
645 new_size
= PAGE_ALIGN(offset
+ size
);
646 if (new_size
< le64toh(f
->header
->header_size
))
647 new_size
= le64toh(f
->header
->header_size
);
649 if (new_size
<= old_size
) {
651 /* We already pre-allocated enough space, but before
652 * we write to it, let's check with fstat() if the
653 * file got deleted, in order make sure we don't throw
654 * away the data immediately. Don't check fstat() for
655 * all writes though, but only once ever 10s. */
657 if (f
->last_stat_usec
+ LAST_STAT_REFRESH_USEC
> now(CLOCK_MONOTONIC
))
660 return journal_file_fstat(f
);
663 /* Allocate more space. */
665 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
668 if (new_size
> f
->metrics
.min_size
&& f
->metrics
.keep_free
> 0) {
671 if (fstatvfs(f
->fd
, &svfs
) >= 0) {
674 available
= LESS_BY((uint64_t) svfs
.f_bfree
* (uint64_t) svfs
.f_bsize
, f
->metrics
.keep_free
);
676 if (new_size
- old_size
> available
)
681 /* Increase by larger blocks at once */
682 new_size
= ((new_size
+FILE_SIZE_INCREASE
-1) / FILE_SIZE_INCREASE
) * FILE_SIZE_INCREASE
;
683 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
684 new_size
= f
->metrics
.max_size
;
686 /* Note that the glibc fallocate() fallback is very
687 inefficient, hence we try to minimize the allocation area
689 r
= posix_fallocate(f
->fd
, old_size
, new_size
- old_size
);
693 f
->header
->arena_size
= htole64(new_size
- le64toh(f
->header
->header_size
));
695 return journal_file_fstat(f
);
698 static unsigned type_to_context(ObjectType type
) {
699 /* One context for each type, plus one catch-all for the rest */
700 assert_cc(_OBJECT_TYPE_MAX
<= MMAP_CACHE_MAX_CONTEXTS
);
701 assert_cc(CONTEXT_HEADER
< MMAP_CACHE_MAX_CONTEXTS
);
702 return type
> OBJECT_UNUSED
&& type
< _OBJECT_TYPE_MAX
? type
: 0;
705 static int journal_file_move_to(JournalFile
*f
, ObjectType type
, bool keep_always
, uint64_t offset
, uint64_t size
, void **ret
) {
714 /* Avoid SIGBUS on invalid accesses */
715 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
) {
716 /* Hmm, out of range? Let's refresh the fstat() data
717 * first, before we trust that check. */
719 r
= journal_file_fstat(f
);
723 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
)
724 return -EADDRNOTAVAIL
;
727 return mmap_cache_get(f
->mmap
, f
->fd
, f
->prot
, type_to_context(type
), keep_always
, offset
, size
, &f
->last_stat
, ret
);
730 static uint64_t minimum_header_size(Object
*o
) {
732 static const uint64_t table
[] = {
733 [OBJECT_DATA
] = sizeof(DataObject
),
734 [OBJECT_FIELD
] = sizeof(FieldObject
),
735 [OBJECT_ENTRY
] = sizeof(EntryObject
),
736 [OBJECT_DATA_HASH_TABLE
] = sizeof(HashTableObject
),
737 [OBJECT_FIELD_HASH_TABLE
] = sizeof(HashTableObject
),
738 [OBJECT_ENTRY_ARRAY
] = sizeof(EntryArrayObject
),
739 [OBJECT_TAG
] = sizeof(TagObject
),
742 if (o
->object
.type
>= ELEMENTSOF(table
) || table
[o
->object
.type
] <= 0)
743 return sizeof(ObjectHeader
);
745 return table
[o
->object
.type
];
748 int journal_file_move_to_object(JournalFile
*f
, ObjectType type
, uint64_t offset
, Object
**ret
) {
757 /* Objects may only be located at multiple of 64 bit */
758 if (!VALID64(offset
)) {
759 log_debug("Attempt to move to object at non-64bit boundary: %" PRIu64
, offset
);
763 /* Object may not be located in the file header */
764 if (offset
< le64toh(f
->header
->header_size
)) {
765 log_debug("Attempt to move to object located in file header: %" PRIu64
, offset
);
769 r
= journal_file_move_to(f
, type
, false, offset
, sizeof(ObjectHeader
), &t
);
774 s
= le64toh(o
->object
.size
);
777 log_debug("Attempt to move to uninitialized object: %" PRIu64
, offset
);
780 if (s
< sizeof(ObjectHeader
)) {
781 log_debug("Attempt to move to overly short object: %" PRIu64
, offset
);
785 if (o
->object
.type
<= OBJECT_UNUSED
) {
786 log_debug("Attempt to move to object with invalid type: %" PRIu64
, offset
);
790 if (s
< minimum_header_size(o
)) {
791 log_debug("Attempt to move to truncated object: %" PRIu64
, offset
);
795 if (type
> OBJECT_UNUSED
&& o
->object
.type
!= type
) {
796 log_debug("Attempt to move to object of unexpected type: %" PRIu64
, offset
);
800 if (s
> sizeof(ObjectHeader
)) {
801 r
= journal_file_move_to(f
, type
, false, offset
, s
, &t
);
812 static uint64_t journal_file_entry_seqnum(JournalFile
*f
, uint64_t *seqnum
) {
818 r
= le64toh(f
->header
->tail_entry_seqnum
) + 1;
821 /* If an external seqnum counter was passed, we update
822 * both the local and the external one, and set it to
823 * the maximum of both */
831 f
->header
->tail_entry_seqnum
= htole64(r
);
833 if (f
->header
->head_entry_seqnum
== 0)
834 f
->header
->head_entry_seqnum
= htole64(r
);
839 int journal_file_append_object(JournalFile
*f
, ObjectType type
, uint64_t size
, Object
**ret
, uint64_t *offset
) {
847 assert(type
> OBJECT_UNUSED
&& type
< _OBJECT_TYPE_MAX
);
848 assert(size
>= sizeof(ObjectHeader
));
852 r
= journal_file_set_online(f
);
856 p
= le64toh(f
->header
->tail_object_offset
);
858 p
= le64toh(f
->header
->header_size
);
860 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &tail
);
864 p
+= ALIGN64(le64toh(tail
->object
.size
));
867 r
= journal_file_allocate(f
, p
, size
);
871 r
= journal_file_move_to(f
, type
, false, p
, size
, &t
);
878 o
->object
.type
= type
;
879 o
->object
.size
= htole64(size
);
881 f
->header
->tail_object_offset
= htole64(p
);
882 f
->header
->n_objects
= htole64(le64toh(f
->header
->n_objects
) + 1);
890 static int journal_file_setup_data_hash_table(JournalFile
*f
) {
898 /* We estimate that we need 1 hash table entry per 768 bytes
899 of journal file and we want to make sure we never get
900 beyond 75% fill level. Calculate the hash table size for
901 the maximum file size based on these metrics. */
903 s
= (f
->metrics
.max_size
* 4 / 768 / 3) * sizeof(HashItem
);
904 if (s
< DEFAULT_DATA_HASH_TABLE_SIZE
)
905 s
= DEFAULT_DATA_HASH_TABLE_SIZE
;
907 log_debug("Reserving %"PRIu64
" entries in hash table.", s
/ sizeof(HashItem
));
909 r
= journal_file_append_object(f
,
910 OBJECT_DATA_HASH_TABLE
,
911 offsetof(Object
, hash_table
.items
) + s
,
916 memzero(o
->hash_table
.items
, s
);
918 f
->header
->data_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
919 f
->header
->data_hash_table_size
= htole64(s
);
924 static int journal_file_setup_field_hash_table(JournalFile
*f
) {
932 /* We use a fixed size hash table for the fields as this
933 * number should grow very slowly only */
935 s
= DEFAULT_FIELD_HASH_TABLE_SIZE
;
936 r
= journal_file_append_object(f
,
937 OBJECT_FIELD_HASH_TABLE
,
938 offsetof(Object
, hash_table
.items
) + s
,
943 memzero(o
->hash_table
.items
, s
);
945 f
->header
->field_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
946 f
->header
->field_hash_table_size
= htole64(s
);
951 int journal_file_map_data_hash_table(JournalFile
*f
) {
959 if (f
->data_hash_table
)
962 p
= le64toh(f
->header
->data_hash_table_offset
);
963 s
= le64toh(f
->header
->data_hash_table_size
);
965 r
= journal_file_move_to(f
,
966 OBJECT_DATA_HASH_TABLE
,
973 f
->data_hash_table
= t
;
977 int journal_file_map_field_hash_table(JournalFile
*f
) {
985 if (f
->field_hash_table
)
988 p
= le64toh(f
->header
->field_hash_table_offset
);
989 s
= le64toh(f
->header
->field_hash_table_size
);
991 r
= journal_file_move_to(f
,
992 OBJECT_FIELD_HASH_TABLE
,
999 f
->field_hash_table
= t
;
1003 static int journal_file_link_field(
1014 assert(f
->field_hash_table
);
1018 if (o
->object
.type
!= OBJECT_FIELD
)
1021 m
= le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
);
1025 /* This might alter the window we are looking at */
1026 o
->field
.next_hash_offset
= o
->field
.head_data_offset
= 0;
1029 p
= le64toh(f
->field_hash_table
[h
].tail_hash_offset
);
1031 f
->field_hash_table
[h
].head_hash_offset
= htole64(offset
);
1033 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1037 o
->field
.next_hash_offset
= htole64(offset
);
1040 f
->field_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1042 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
1043 f
->header
->n_fields
= htole64(le64toh(f
->header
->n_fields
) + 1);
1048 static int journal_file_link_data(
1059 assert(f
->data_hash_table
);
1063 if (o
->object
.type
!= OBJECT_DATA
)
1066 m
= le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
);
1070 /* This might alter the window we are looking at */
1071 o
->data
.next_hash_offset
= o
->data
.next_field_offset
= 0;
1072 o
->data
.entry_offset
= o
->data
.entry_array_offset
= 0;
1073 o
->data
.n_entries
= 0;
1076 p
= le64toh(f
->data_hash_table
[h
].tail_hash_offset
);
1078 /* Only entry in the hash table is easy */
1079 f
->data_hash_table
[h
].head_hash_offset
= htole64(offset
);
1081 /* Move back to the previous data object, to patch in
1084 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1088 o
->data
.next_hash_offset
= htole64(offset
);
1091 f
->data_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1093 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
1094 f
->header
->n_data
= htole64(le64toh(f
->header
->n_data
) + 1);
1099 int journal_file_find_field_object_with_hash(
1101 const void *field
, uint64_t size
, uint64_t hash
,
1102 Object
**ret
, uint64_t *offset
) {
1104 uint64_t p
, osize
, h
, m
;
1109 assert(field
&& size
> 0);
1111 /* If the field hash table is empty, we can't find anything */
1112 if (le64toh(f
->header
->field_hash_table_size
) <= 0)
1115 /* Map the field hash table, if it isn't mapped yet. */
1116 r
= journal_file_map_field_hash_table(f
);
1120 osize
= offsetof(Object
, field
.payload
) + size
;
1122 m
= le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
);
1127 p
= le64toh(f
->field_hash_table
[h
].head_hash_offset
);
1132 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1136 if (le64toh(o
->field
.hash
) == hash
&&
1137 le64toh(o
->object
.size
) == osize
&&
1138 memcmp(o
->field
.payload
, field
, size
) == 0) {
1148 p
= le64toh(o
->field
.next_hash_offset
);
1154 int journal_file_find_field_object(
1156 const void *field
, uint64_t size
,
1157 Object
**ret
, uint64_t *offset
) {
1162 assert(field
&& size
> 0);
1164 hash
= hash64(field
, size
);
1166 return journal_file_find_field_object_with_hash(f
,
1171 int journal_file_find_data_object_with_hash(
1173 const void *data
, uint64_t size
, uint64_t hash
,
1174 Object
**ret
, uint64_t *offset
) {
1176 uint64_t p
, osize
, h
, m
;
1181 assert(data
|| size
== 0);
1183 /* If there's no data hash table, then there's no entry. */
1184 if (le64toh(f
->header
->data_hash_table_size
) <= 0)
1187 /* Map the data hash table, if it isn't mapped yet. */
1188 r
= journal_file_map_data_hash_table(f
);
1192 osize
= offsetof(Object
, data
.payload
) + size
;
1194 m
= le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
);
1199 p
= le64toh(f
->data_hash_table
[h
].head_hash_offset
);
1204 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1208 if (le64toh(o
->data
.hash
) != hash
)
1211 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
) {
1212 #if defined(HAVE_XZ) || defined(HAVE_LZ4)
1216 l
= le64toh(o
->object
.size
);
1217 if (l
<= offsetof(Object
, data
.payload
))
1220 l
-= offsetof(Object
, data
.payload
);
1222 r
= decompress_blob(o
->object
.flags
& OBJECT_COMPRESSION_MASK
,
1223 o
->data
.payload
, l
, &f
->compress_buffer
, &f
->compress_buffer_size
, &rsize
, 0);
1227 if (rsize
== size
&&
1228 memcmp(f
->compress_buffer
, data
, size
) == 0) {
1239 return -EPROTONOSUPPORT
;
1241 } else if (le64toh(o
->object
.size
) == osize
&&
1242 memcmp(o
->data
.payload
, data
, size
) == 0) {
1254 p
= le64toh(o
->data
.next_hash_offset
);
1260 int journal_file_find_data_object(
1262 const void *data
, uint64_t size
,
1263 Object
**ret
, uint64_t *offset
) {
1268 assert(data
|| size
== 0);
1270 hash
= hash64(data
, size
);
1272 return journal_file_find_data_object_with_hash(f
,
1277 static int journal_file_append_field(
1279 const void *field
, uint64_t size
,
1280 Object
**ret
, uint64_t *offset
) {
1288 assert(field
&& size
> 0);
1290 hash
= hash64(field
, size
);
1292 r
= journal_file_find_field_object_with_hash(f
, field
, size
, hash
, &o
, &p
);
1306 osize
= offsetof(Object
, field
.payload
) + size
;
1307 r
= journal_file_append_object(f
, OBJECT_FIELD
, osize
, &o
, &p
);
1311 o
->field
.hash
= htole64(hash
);
1312 memcpy(o
->field
.payload
, field
, size
);
1314 r
= journal_file_link_field(f
, o
, p
, hash
);
1318 /* The linking might have altered the window, so let's
1319 * refresh our pointer */
1320 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1325 r
= journal_file_hmac_put_object(f
, OBJECT_FIELD
, o
, p
);
1339 static int journal_file_append_data(
1341 const void *data
, uint64_t size
,
1342 Object
**ret
, uint64_t *offset
) {
1347 int r
, compression
= 0;
1351 assert(data
|| size
== 0);
1353 hash
= hash64(data
, size
);
1355 r
= journal_file_find_data_object_with_hash(f
, data
, size
, hash
, &o
, &p
);
1369 osize
= offsetof(Object
, data
.payload
) + size
;
1370 r
= journal_file_append_object(f
, OBJECT_DATA
, osize
, &o
, &p
);
1374 o
->data
.hash
= htole64(hash
);
1376 #if defined(HAVE_XZ) || defined(HAVE_LZ4)
1377 if (JOURNAL_FILE_COMPRESS(f
) && size
>= COMPRESSION_SIZE_THRESHOLD
) {
1380 compression
= compress_blob(data
, size
, o
->data
.payload
, size
- 1, &rsize
);
1382 if (compression
>= 0) {
1383 o
->object
.size
= htole64(offsetof(Object
, data
.payload
) + rsize
);
1384 o
->object
.flags
|= compression
;
1386 log_debug("Compressed data object %"PRIu64
" -> %zu using %s",
1387 size
, rsize
, object_compressed_to_string(compression
));
1389 /* Compression didn't work, we don't really care why, let's continue without compression */
1394 if (compression
== 0)
1395 memcpy_safe(o
->data
.payload
, data
, size
);
1397 r
= journal_file_link_data(f
, o
, p
, hash
);
1402 r
= journal_file_hmac_put_object(f
, OBJECT_DATA
, o
, p
);
1407 /* The linking might have altered the window, so let's
1408 * refresh our pointer */
1409 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1416 eq
= memchr(data
, '=', size
);
1417 if (eq
&& eq
> data
) {
1421 /* Create field object ... */
1422 r
= journal_file_append_field(f
, data
, (uint8_t*) eq
- (uint8_t*) data
, &fo
, &fp
);
1426 /* ... and link it in. */
1427 o
->data
.next_field_offset
= fo
->field
.head_data_offset
;
1428 fo
->field
.head_data_offset
= le64toh(p
);
1440 uint64_t journal_file_entry_n_items(Object
*o
) {
1443 if (o
->object
.type
!= OBJECT_ENTRY
)
1446 return (le64toh(o
->object
.size
) - offsetof(Object
, entry
.items
)) / sizeof(EntryItem
);
1449 uint64_t journal_file_entry_array_n_items(Object
*o
) {
1452 if (o
->object
.type
!= OBJECT_ENTRY_ARRAY
)
1455 return (le64toh(o
->object
.size
) - offsetof(Object
, entry_array
.items
)) / sizeof(uint64_t);
1458 uint64_t journal_file_hash_table_n_items(Object
*o
) {
1461 if (o
->object
.type
!= OBJECT_DATA_HASH_TABLE
&&
1462 o
->object
.type
!= OBJECT_FIELD_HASH_TABLE
)
1465 return (le64toh(o
->object
.size
) - offsetof(Object
, hash_table
.items
)) / sizeof(HashItem
);
1468 static int link_entry_into_array(JournalFile
*f
,
1473 uint64_t n
= 0, ap
= 0, q
, i
, a
, hidx
;
1482 a
= le64toh(*first
);
1483 i
= hidx
= le64toh(*idx
);
1486 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
1490 n
= journal_file_entry_array_n_items(o
);
1492 o
->entry_array
.items
[i
] = htole64(p
);
1493 *idx
= htole64(hidx
+ 1);
1499 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
1510 r
= journal_file_append_object(f
, OBJECT_ENTRY_ARRAY
,
1511 offsetof(Object
, entry_array
.items
) + n
* sizeof(uint64_t),
1517 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY_ARRAY
, o
, q
);
1522 o
->entry_array
.items
[i
] = htole64(p
);
1525 *first
= htole64(q
);
1527 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, ap
, &o
);
1531 o
->entry_array
.next_entry_array_offset
= htole64(q
);
1534 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
1535 f
->header
->n_entry_arrays
= htole64(le64toh(f
->header
->n_entry_arrays
) + 1);
1537 *idx
= htole64(hidx
+ 1);
1542 static int link_entry_into_array_plus_one(JournalFile
*f
,
1557 *extra
= htole64(p
);
1561 i
= htole64(le64toh(*idx
) - 1);
1562 r
= link_entry_into_array(f
, first
, &i
, p
);
1567 *idx
= htole64(le64toh(*idx
) + 1);
1571 static int journal_file_link_entry_item(JournalFile
*f
, Object
*o
, uint64_t offset
, uint64_t i
) {
1578 p
= le64toh(o
->entry
.items
[i
].object_offset
);
1582 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1586 return link_entry_into_array_plus_one(f
,
1587 &o
->data
.entry_offset
,
1588 &o
->data
.entry_array_offset
,
1593 static int journal_file_link_entry(JournalFile
*f
, Object
*o
, uint64_t offset
) {
1602 if (o
->object
.type
!= OBJECT_ENTRY
)
1605 __sync_synchronize();
1607 /* Link up the entry itself */
1608 r
= link_entry_into_array(f
,
1609 &f
->header
->entry_array_offset
,
1610 &f
->header
->n_entries
,
1615 /* log_debug("=> %s seqnr=%"PRIu64" n_entries=%"PRIu64, f->path, o->entry.seqnum, f->header->n_entries); */
1617 if (f
->header
->head_entry_realtime
== 0)
1618 f
->header
->head_entry_realtime
= o
->entry
.realtime
;
1620 f
->header
->tail_entry_realtime
= o
->entry
.realtime
;
1621 f
->header
->tail_entry_monotonic
= o
->entry
.monotonic
;
1623 f
->tail_entry_monotonic_valid
= true;
1625 /* Link up the items */
1626 n
= journal_file_entry_n_items(o
);
1627 for (i
= 0; i
< n
; i
++) {
1628 r
= journal_file_link_entry_item(f
, o
, offset
, i
);
1636 static int journal_file_append_entry_internal(
1638 const dual_timestamp
*ts
,
1640 const EntryItem items
[], unsigned n_items
,
1642 Object
**ret
, uint64_t *offset
) {
1650 assert(items
|| n_items
== 0);
1653 osize
= offsetof(Object
, entry
.items
) + (n_items
* sizeof(EntryItem
));
1655 r
= journal_file_append_object(f
, OBJECT_ENTRY
, osize
, &o
, &np
);
1659 o
->entry
.seqnum
= htole64(journal_file_entry_seqnum(f
, seqnum
));
1660 memcpy_safe(o
->entry
.items
, items
, n_items
* sizeof(EntryItem
));
1661 o
->entry
.realtime
= htole64(ts
->realtime
);
1662 o
->entry
.monotonic
= htole64(ts
->monotonic
);
1663 o
->entry
.xor_hash
= htole64(xor_hash
);
1664 o
->entry
.boot_id
= f
->header
->boot_id
;
1667 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY
, o
, np
);
1672 r
= journal_file_link_entry(f
, o
, np
);
1685 void journal_file_post_change(JournalFile
*f
) {
1688 /* inotify() does not receive IN_MODIFY events from file
1689 * accesses done via mmap(). After each access we hence
1690 * trigger IN_MODIFY by truncating the journal file to its
1691 * current size which triggers IN_MODIFY. */
1693 __sync_synchronize();
1695 if (ftruncate(f
->fd
, f
->last_stat
.st_size
) < 0)
1696 log_debug_errno(errno
, "Failed to truncate file to its own size: %m");
1699 static int post_change_thunk(sd_event_source
*timer
, uint64_t usec
, void *userdata
) {
1702 journal_file_post_change(userdata
);
1707 static void schedule_post_change(JournalFile
*f
) {
1708 sd_event_source
*timer
;
1713 assert(f
->post_change_timer
);
1715 timer
= f
->post_change_timer
;
1717 r
= sd_event_source_get_enabled(timer
, &enabled
);
1719 log_debug_errno(r
, "Failed to get ftruncate timer state: %m");
1723 if (enabled
== SD_EVENT_ONESHOT
)
1726 r
= sd_event_now(sd_event_source_get_event(timer
), CLOCK_MONOTONIC
, &now
);
1728 log_debug_errno(r
, "Failed to get clock's now for scheduling ftruncate: %m");
1732 r
= sd_event_source_set_time(timer
, now
+f
->post_change_timer_period
);
1734 log_debug_errno(r
, "Failed to set time for scheduling ftruncate: %m");
1738 r
= sd_event_source_set_enabled(timer
, SD_EVENT_ONESHOT
);
1740 log_debug_errno(r
, "Failed to enable scheduled ftruncate: %m");
1747 /* On failure, let's simply post the change immediately. */
1748 journal_file_post_change(f
);
1751 /* Enable coalesced change posting in a timer on the provided sd_event instance */
1752 int journal_file_enable_post_change_timer(JournalFile
*f
, sd_event
*e
, usec_t t
) {
1753 _cleanup_(sd_event_source_unrefp
) sd_event_source
*timer
= NULL
;
1757 assert_return(!f
->post_change_timer
, -EINVAL
);
1761 r
= sd_event_add_time(e
, &timer
, CLOCK_MONOTONIC
, 0, 0, post_change_thunk
, f
);
1765 r
= sd_event_source_set_enabled(timer
, SD_EVENT_OFF
);
1769 f
->post_change_timer
= timer
;
1771 f
->post_change_timer_period
= t
;
1776 static int entry_item_cmp(const void *_a
, const void *_b
) {
1777 const EntryItem
*a
= _a
, *b
= _b
;
1779 if (le64toh(a
->object_offset
) < le64toh(b
->object_offset
))
1781 if (le64toh(a
->object_offset
) > le64toh(b
->object_offset
))
1786 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
) {
1790 uint64_t xor_hash
= 0;
1791 struct dual_timestamp _ts
;
1795 assert(iovec
|| n_iovec
== 0);
1798 dual_timestamp_get(&_ts
);
1803 r
= journal_file_maybe_append_tag(f
, ts
->realtime
);
1808 /* alloca() can't take 0, hence let's allocate at least one */
1809 items
= alloca(sizeof(EntryItem
) * MAX(1u, n_iovec
));
1811 for (i
= 0; i
< n_iovec
; i
++) {
1815 r
= journal_file_append_data(f
, iovec
[i
].iov_base
, iovec
[i
].iov_len
, &o
, &p
);
1819 xor_hash
^= le64toh(o
->data
.hash
);
1820 items
[i
].object_offset
= htole64(p
);
1821 items
[i
].hash
= o
->data
.hash
;
1824 /* Order by the position on disk, in order to improve seek
1825 * times for rotating media. */
1826 qsort_safe(items
, n_iovec
, sizeof(EntryItem
), entry_item_cmp
);
1828 r
= journal_file_append_entry_internal(f
, ts
, xor_hash
, items
, n_iovec
, seqnum
, ret
, offset
);
1830 /* If the memory mapping triggered a SIGBUS then we return an
1831 * IO error and ignore the error code passed down to us, since
1832 * it is very likely just an effect of a nullified replacement
1835 if (mmap_cache_got_sigbus(f
->mmap
, f
->fd
))
1838 if (f
->post_change_timer
)
1839 schedule_post_change(f
);
1841 journal_file_post_change(f
);
1846 typedef struct ChainCacheItem
{
1847 uint64_t first
; /* the array at the beginning of the chain */
1848 uint64_t array
; /* the cached array */
1849 uint64_t begin
; /* the first item in the cached array */
1850 uint64_t total
; /* the total number of items in all arrays before this one in the chain */
1851 uint64_t last_index
; /* the last index we looked at, to optimize locality when bisecting */
1854 static void chain_cache_put(
1861 uint64_t last_index
) {
1864 /* If the chain item to cache for this chain is the
1865 * first one it's not worth caching anything */
1869 if (ordered_hashmap_size(h
) >= CHAIN_CACHE_MAX
) {
1870 ci
= ordered_hashmap_steal_first(h
);
1873 ci
= new(ChainCacheItem
, 1);
1880 if (ordered_hashmap_put(h
, &ci
->first
, ci
) < 0) {
1885 assert(ci
->first
== first
);
1890 ci
->last_index
= last_index
;
1893 static int generic_array_get(
1897 Object
**ret
, uint64_t *offset
) {
1900 uint64_t p
= 0, a
, t
= 0;
1908 /* Try the chain cache first */
1909 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
1910 if (ci
&& i
> ci
->total
) {
1919 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
1923 k
= journal_file_entry_array_n_items(o
);
1925 p
= le64toh(o
->entry_array
.items
[i
]);
1931 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
1937 /* Let's cache this item for the next invocation */
1938 chain_cache_put(f
->chain_cache
, ci
, first
, a
, le64toh(o
->entry_array
.items
[0]), t
, i
);
1940 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
1953 static int generic_array_get_plus_one(
1958 Object
**ret
, uint64_t *offset
) {
1967 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, &o
);
1980 return generic_array_get(f
, first
, i
-1, ret
, offset
);
1989 static int generic_array_bisect(
1994 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
1995 direction_t direction
,
2000 uint64_t a
, p
, t
= 0, i
= 0, last_p
= 0, last_index
= (uint64_t) -1;
2001 bool subtract_one
= false;
2002 Object
*o
, *array
= NULL
;
2007 assert(test_object
);
2009 /* Start with the first array in the chain */
2012 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
2013 if (ci
&& n
> ci
->total
) {
2014 /* Ah, we have iterated this bisection array chain
2015 * previously! Let's see if we can skip ahead in the
2016 * chain, as far as the last time. But we can't jump
2017 * backwards in the chain, so let's check that
2020 r
= test_object(f
, ci
->begin
, needle
);
2024 if (r
== TEST_LEFT
) {
2025 /* OK, what we are looking for is right of the
2026 * begin of this EntryArray, so let's jump
2027 * straight to previously cached array in the
2033 last_index
= ci
->last_index
;
2038 uint64_t left
, right
, k
, lp
;
2040 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &array
);
2044 k
= journal_file_entry_array_n_items(array
);
2050 lp
= p
= le64toh(array
->entry_array
.items
[i
]);
2054 r
= test_object(f
, p
, needle
);
2055 if (r
== -EBADMSG
) {
2056 log_debug_errno(r
, "Encountered invalid entry while bisecting, cutting algorithm short. (1)");
2063 if (r
== TEST_FOUND
)
2064 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2066 if (r
== TEST_RIGHT
) {
2070 if (last_index
!= (uint64_t) -1) {
2071 assert(last_index
<= right
);
2073 /* If we cached the last index we
2074 * looked at, let's try to not to jump
2075 * too wildly around and see if we can
2076 * limit the range to look at early to
2077 * the immediate neighbors of the last
2078 * index we looked at. */
2080 if (last_index
> 0) {
2081 uint64_t x
= last_index
- 1;
2083 p
= le64toh(array
->entry_array
.items
[x
]);
2087 r
= test_object(f
, p
, needle
);
2091 if (r
== TEST_FOUND
)
2092 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2094 if (r
== TEST_RIGHT
)
2100 if (last_index
< right
) {
2101 uint64_t y
= last_index
+ 1;
2103 p
= le64toh(array
->entry_array
.items
[y
]);
2107 r
= test_object(f
, p
, needle
);
2111 if (r
== TEST_FOUND
)
2112 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2114 if (r
== TEST_RIGHT
)
2122 if (left
== right
) {
2123 if (direction
== DIRECTION_UP
)
2124 subtract_one
= true;
2130 assert(left
< right
);
2131 i
= (left
+ right
) / 2;
2133 p
= le64toh(array
->entry_array
.items
[i
]);
2137 r
= test_object(f
, p
, needle
);
2138 if (r
== -EBADMSG
) {
2139 log_debug_errno(r
, "Encountered invalid entry while bisecting, cutting algorithm short. (2)");
2146 if (r
== TEST_FOUND
)
2147 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2149 if (r
== TEST_RIGHT
)
2157 if (direction
== DIRECTION_UP
) {
2159 subtract_one
= true;
2170 last_index
= (uint64_t) -1;
2171 a
= le64toh(array
->entry_array
.next_entry_array_offset
);
2177 if (subtract_one
&& t
== 0 && i
== 0)
2180 /* Let's cache this item for the next invocation */
2181 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
);
2183 if (subtract_one
&& i
== 0)
2185 else if (subtract_one
)
2186 p
= le64toh(array
->entry_array
.items
[i
-1]);
2188 p
= le64toh(array
->entry_array
.items
[i
]);
2190 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2201 *idx
= t
+ i
+ (subtract_one
? -1 : 0);
2206 static int generic_array_bisect_plus_one(
2212 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
2213 direction_t direction
,
2219 bool step_back
= false;
2223 assert(test_object
);
2228 /* This bisects the array in object 'first', but first checks
2230 r
= test_object(f
, extra
, needle
);
2234 if (r
== TEST_FOUND
)
2235 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2237 /* if we are looking with DIRECTION_UP then we need to first
2238 see if in the actual array there is a matching entry, and
2239 return the last one of that. But if there isn't any we need
2240 to return this one. Hence remember this, and return it
2243 step_back
= direction
== DIRECTION_UP
;
2245 if (r
== TEST_RIGHT
) {
2246 if (direction
== DIRECTION_DOWN
)
2252 r
= generic_array_bisect(f
, first
, n
-1, needle
, test_object
, direction
, ret
, offset
, idx
);
2254 if (r
== 0 && step_back
)
2263 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, &o
);
2279 _pure_
static int test_object_offset(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2285 else if (p
< needle
)
2291 static int test_object_seqnum(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2298 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2302 if (le64toh(o
->entry
.seqnum
) == needle
)
2304 else if (le64toh(o
->entry
.seqnum
) < needle
)
2310 int journal_file_move_to_entry_by_seqnum(
2313 direction_t direction
,
2319 return generic_array_bisect(f
,
2320 le64toh(f
->header
->entry_array_offset
),
2321 le64toh(f
->header
->n_entries
),
2328 static int test_object_realtime(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2335 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2339 if (le64toh(o
->entry
.realtime
) == needle
)
2341 else if (le64toh(o
->entry
.realtime
) < needle
)
2347 int journal_file_move_to_entry_by_realtime(
2350 direction_t direction
,
2356 return generic_array_bisect(f
,
2357 le64toh(f
->header
->entry_array_offset
),
2358 le64toh(f
->header
->n_entries
),
2360 test_object_realtime
,
2365 static int test_object_monotonic(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2372 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2376 if (le64toh(o
->entry
.monotonic
) == needle
)
2378 else if (le64toh(o
->entry
.monotonic
) < needle
)
2384 static int find_data_object_by_boot_id(
2390 char t
[sizeof("_BOOT_ID=")-1 + 32 + 1] = "_BOOT_ID=";
2392 sd_id128_to_string(boot_id
, t
+ 9);
2393 return journal_file_find_data_object(f
, t
, sizeof(t
) - 1, o
, b
);
2396 int journal_file_move_to_entry_by_monotonic(
2400 direction_t direction
,
2409 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, NULL
);
2415 return generic_array_bisect_plus_one(f
,
2416 le64toh(o
->data
.entry_offset
),
2417 le64toh(o
->data
.entry_array_offset
),
2418 le64toh(o
->data
.n_entries
),
2420 test_object_monotonic
,
2425 void journal_file_reset_location(JournalFile
*f
) {
2426 f
->location_type
= LOCATION_HEAD
;
2427 f
->current_offset
= 0;
2428 f
->current_seqnum
= 0;
2429 f
->current_realtime
= 0;
2430 f
->current_monotonic
= 0;
2431 zero(f
->current_boot_id
);
2432 f
->current_xor_hash
= 0;
2435 void journal_file_save_location(JournalFile
*f
, Object
*o
, uint64_t offset
) {
2436 f
->location_type
= LOCATION_SEEK
;
2437 f
->current_offset
= offset
;
2438 f
->current_seqnum
= le64toh(o
->entry
.seqnum
);
2439 f
->current_realtime
= le64toh(o
->entry
.realtime
);
2440 f
->current_monotonic
= le64toh(o
->entry
.monotonic
);
2441 f
->current_boot_id
= o
->entry
.boot_id
;
2442 f
->current_xor_hash
= le64toh(o
->entry
.xor_hash
);
2445 int journal_file_compare_locations(JournalFile
*af
, JournalFile
*bf
) {
2450 assert(af
->location_type
== LOCATION_SEEK
);
2451 assert(bf
->location_type
== LOCATION_SEEK
);
2453 /* If contents and timestamps match, these entries are
2454 * identical, even if the seqnum does not match */
2455 if (sd_id128_equal(af
->current_boot_id
, bf
->current_boot_id
) &&
2456 af
->current_monotonic
== bf
->current_monotonic
&&
2457 af
->current_realtime
== bf
->current_realtime
&&
2458 af
->current_xor_hash
== bf
->current_xor_hash
)
2461 if (sd_id128_equal(af
->header
->seqnum_id
, bf
->header
->seqnum_id
)) {
2463 /* If this is from the same seqnum source, compare
2465 if (af
->current_seqnum
< bf
->current_seqnum
)
2467 if (af
->current_seqnum
> bf
->current_seqnum
)
2470 /* Wow! This is weird, different data but the same
2471 * seqnums? Something is borked, but let's make the
2472 * best of it and compare by time. */
2475 if (sd_id128_equal(af
->current_boot_id
, bf
->current_boot_id
)) {
2477 /* If the boot id matches, compare monotonic time */
2478 if (af
->current_monotonic
< bf
->current_monotonic
)
2480 if (af
->current_monotonic
> bf
->current_monotonic
)
2484 /* Otherwise, compare UTC time */
2485 if (af
->current_realtime
< bf
->current_realtime
)
2487 if (af
->current_realtime
> bf
->current_realtime
)
2490 /* Finally, compare by contents */
2491 if (af
->current_xor_hash
< bf
->current_xor_hash
)
2493 if (af
->current_xor_hash
> bf
->current_xor_hash
)
2499 static int bump_array_index(uint64_t *i
, direction_t direction
, uint64_t n
) {
2501 /* Increase or decrease the specified index, in the right direction. */
2503 if (direction
== DIRECTION_DOWN
) {
2518 static bool check_properly_ordered(uint64_t new_offset
, uint64_t old_offset
, direction_t direction
) {
2520 /* Consider it an error if any of the two offsets is uninitialized */
2521 if (old_offset
== 0 || new_offset
== 0)
2524 /* If we go down, the new offset must be larger than the old one. */
2525 return direction
== DIRECTION_DOWN
?
2526 new_offset
> old_offset
:
2527 new_offset
< old_offset
;
2530 int journal_file_next_entry(
2533 direction_t direction
,
2534 Object
**ret
, uint64_t *offset
) {
2542 n
= le64toh(f
->header
->n_entries
);
2547 i
= direction
== DIRECTION_DOWN
? 0 : n
- 1;
2549 r
= generic_array_bisect(f
,
2550 le64toh(f
->header
->entry_array_offset
),
2551 le64toh(f
->header
->n_entries
),
2560 r
= bump_array_index(&i
, direction
, n
);
2565 /* And jump to it */
2567 r
= generic_array_get(f
,
2568 le64toh(f
->header
->entry_array_offset
),
2576 /* OK, so this entry is borked. Most likely some entry didn't get synced to disk properly, let's see if
2577 * the next one might work for us instead. */
2578 log_debug_errno(r
, "Entry item %" PRIu64
" is bad, skipping over it.", i
);
2580 r
= bump_array_index(&i
, direction
, n
);
2585 /* Ensure our array is properly ordered. */
2586 if (p
> 0 && !check_properly_ordered(ofs
, p
, direction
)) {
2587 log_debug("%s: entry array not properly ordered at entry %" PRIu64
, f
->path
, i
);
2597 int journal_file_next_entry_for_data(
2599 Object
*o
, uint64_t p
,
2600 uint64_t data_offset
,
2601 direction_t direction
,
2602 Object
**ret
, uint64_t *offset
) {
2609 assert(p
> 0 || !o
);
2611 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2615 n
= le64toh(d
->data
.n_entries
);
2620 i
= direction
== DIRECTION_DOWN
? 0 : n
- 1;
2622 if (o
->object
.type
!= OBJECT_ENTRY
)
2625 r
= generic_array_bisect_plus_one(f
,
2626 le64toh(d
->data
.entry_offset
),
2627 le64toh(d
->data
.entry_array_offset
),
2628 le64toh(d
->data
.n_entries
),
2638 r
= bump_array_index(&i
, direction
, n
);
2644 r
= generic_array_get_plus_one(f
,
2645 le64toh(d
->data
.entry_offset
),
2646 le64toh(d
->data
.entry_array_offset
),
2654 log_debug_errno(r
, "Data entry item %" PRIu64
" is bad, skipping over it.", i
);
2656 r
= bump_array_index(&i
, direction
, n
);
2661 /* Ensure our array is properly ordered. */
2662 if (p
> 0 && check_properly_ordered(ofs
, p
, direction
)) {
2663 log_debug("%s data entry array not properly ordered at entry %" PRIu64
, f
->path
, i
);
2673 int journal_file_move_to_entry_by_offset_for_data(
2675 uint64_t data_offset
,
2677 direction_t direction
,
2678 Object
**ret
, uint64_t *offset
) {
2685 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2689 return generic_array_bisect_plus_one(f
,
2690 le64toh(d
->data
.entry_offset
),
2691 le64toh(d
->data
.entry_array_offset
),
2692 le64toh(d
->data
.n_entries
),
2699 int journal_file_move_to_entry_by_monotonic_for_data(
2701 uint64_t data_offset
,
2704 direction_t direction
,
2705 Object
**ret
, uint64_t *offset
) {
2713 /* First, seek by time */
2714 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, &b
);
2720 r
= generic_array_bisect_plus_one(f
,
2721 le64toh(o
->data
.entry_offset
),
2722 le64toh(o
->data
.entry_array_offset
),
2723 le64toh(o
->data
.n_entries
),
2725 test_object_monotonic
,
2731 /* And now, continue seeking until we find an entry that
2732 * exists in both bisection arrays */
2738 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2742 r
= generic_array_bisect_plus_one(f
,
2743 le64toh(d
->data
.entry_offset
),
2744 le64toh(d
->data
.entry_array_offset
),
2745 le64toh(d
->data
.n_entries
),
2753 r
= journal_file_move_to_object(f
, OBJECT_DATA
, b
, &o
);
2757 r
= generic_array_bisect_plus_one(f
,
2758 le64toh(o
->data
.entry_offset
),
2759 le64toh(o
->data
.entry_array_offset
),
2760 le64toh(o
->data
.n_entries
),
2782 int journal_file_move_to_entry_by_seqnum_for_data(
2784 uint64_t data_offset
,
2786 direction_t direction
,
2787 Object
**ret
, uint64_t *offset
) {
2794 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2798 return generic_array_bisect_plus_one(f
,
2799 le64toh(d
->data
.entry_offset
),
2800 le64toh(d
->data
.entry_array_offset
),
2801 le64toh(d
->data
.n_entries
),
2808 int journal_file_move_to_entry_by_realtime_for_data(
2810 uint64_t data_offset
,
2812 direction_t direction
,
2813 Object
**ret
, uint64_t *offset
) {
2820 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2824 return generic_array_bisect_plus_one(f
,
2825 le64toh(d
->data
.entry_offset
),
2826 le64toh(d
->data
.entry_array_offset
),
2827 le64toh(d
->data
.n_entries
),
2829 test_object_realtime
,
2834 void journal_file_dump(JournalFile
*f
) {
2842 journal_file_print_header(f
);
2844 p
= le64toh(f
->header
->header_size
);
2846 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &o
);
2850 switch (o
->object
.type
) {
2853 printf("Type: OBJECT_UNUSED\n");
2857 printf("Type: OBJECT_DATA\n");
2861 printf("Type: OBJECT_FIELD\n");
2865 printf("Type: OBJECT_ENTRY seqnum=%"PRIu64
" monotonic=%"PRIu64
" realtime=%"PRIu64
"\n",
2866 le64toh(o
->entry
.seqnum
),
2867 le64toh(o
->entry
.monotonic
),
2868 le64toh(o
->entry
.realtime
));
2871 case OBJECT_FIELD_HASH_TABLE
:
2872 printf("Type: OBJECT_FIELD_HASH_TABLE\n");
2875 case OBJECT_DATA_HASH_TABLE
:
2876 printf("Type: OBJECT_DATA_HASH_TABLE\n");
2879 case OBJECT_ENTRY_ARRAY
:
2880 printf("Type: OBJECT_ENTRY_ARRAY\n");
2884 printf("Type: OBJECT_TAG seqnum=%"PRIu64
" epoch=%"PRIu64
"\n",
2885 le64toh(o
->tag
.seqnum
),
2886 le64toh(o
->tag
.epoch
));
2890 printf("Type: unknown (%i)\n", o
->object
.type
);
2894 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
)
2895 printf("Flags: %s\n",
2896 object_compressed_to_string(o
->object
.flags
& OBJECT_COMPRESSION_MASK
));
2898 if (p
== le64toh(f
->header
->tail_object_offset
))
2901 p
= p
+ ALIGN64(le64toh(o
->object
.size
));
2906 log_error("File corrupt");
2909 static const char* format_timestamp_safe(char *buf
, size_t l
, usec_t t
) {
2912 x
= format_timestamp(buf
, l
, t
);
2918 void journal_file_print_header(JournalFile
*f
) {
2919 char a
[33], b
[33], c
[33], d
[33];
2920 char x
[FORMAT_TIMESTAMP_MAX
], y
[FORMAT_TIMESTAMP_MAX
], z
[FORMAT_TIMESTAMP_MAX
];
2922 char bytes
[FORMAT_BYTES_MAX
];
2927 printf("File Path: %s\n"
2931 "Sequential Number ID: %s\n"
2933 "Compatible Flags:%s%s\n"
2934 "Incompatible Flags:%s%s%s\n"
2935 "Header size: %"PRIu64
"\n"
2936 "Arena size: %"PRIu64
"\n"
2937 "Data Hash Table Size: %"PRIu64
"\n"
2938 "Field Hash Table Size: %"PRIu64
"\n"
2939 "Rotate Suggested: %s\n"
2940 "Head Sequential Number: %"PRIu64
" (%"PRIx64
")\n"
2941 "Tail Sequential Number: %"PRIu64
" (%"PRIx64
")\n"
2942 "Head Realtime Timestamp: %s (%"PRIx64
")\n"
2943 "Tail Realtime Timestamp: %s (%"PRIx64
")\n"
2944 "Tail Monotonic Timestamp: %s (%"PRIx64
")\n"
2945 "Objects: %"PRIu64
"\n"
2946 "Entry Objects: %"PRIu64
"\n",
2948 sd_id128_to_string(f
->header
->file_id
, a
),
2949 sd_id128_to_string(f
->header
->machine_id
, b
),
2950 sd_id128_to_string(f
->header
->boot_id
, c
),
2951 sd_id128_to_string(f
->header
->seqnum_id
, d
),
2952 f
->header
->state
== STATE_OFFLINE
? "OFFLINE" :
2953 f
->header
->state
== STATE_ONLINE
? "ONLINE" :
2954 f
->header
->state
== STATE_ARCHIVED
? "ARCHIVED" : "UNKNOWN",
2955 JOURNAL_HEADER_SEALED(f
->header
) ? " SEALED" : "",
2956 (le32toh(f
->header
->compatible_flags
) & ~HEADER_COMPATIBLE_ANY
) ? " ???" : "",
2957 JOURNAL_HEADER_COMPRESSED_XZ(f
->header
) ? " COMPRESSED-XZ" : "",
2958 JOURNAL_HEADER_COMPRESSED_LZ4(f
->header
) ? " COMPRESSED-LZ4" : "",
2959 (le32toh(f
->header
->incompatible_flags
) & ~HEADER_INCOMPATIBLE_ANY
) ? " ???" : "",
2960 le64toh(f
->header
->header_size
),
2961 le64toh(f
->header
->arena_size
),
2962 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
2963 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
),
2964 yes_no(journal_file_rotate_suggested(f
, 0)),
2965 le64toh(f
->header
->head_entry_seqnum
), le64toh(f
->header
->head_entry_seqnum
),
2966 le64toh(f
->header
->tail_entry_seqnum
), le64toh(f
->header
->tail_entry_seqnum
),
2967 format_timestamp_safe(x
, sizeof(x
), le64toh(f
->header
->head_entry_realtime
)), le64toh(f
->header
->head_entry_realtime
),
2968 format_timestamp_safe(y
, sizeof(y
), le64toh(f
->header
->tail_entry_realtime
)), le64toh(f
->header
->tail_entry_realtime
),
2969 format_timespan(z
, sizeof(z
), le64toh(f
->header
->tail_entry_monotonic
), USEC_PER_MSEC
), le64toh(f
->header
->tail_entry_monotonic
),
2970 le64toh(f
->header
->n_objects
),
2971 le64toh(f
->header
->n_entries
));
2973 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
2974 printf("Data Objects: %"PRIu64
"\n"
2975 "Data Hash Table Fill: %.1f%%\n",
2976 le64toh(f
->header
->n_data
),
2977 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))));
2979 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
2980 printf("Field Objects: %"PRIu64
"\n"
2981 "Field Hash Table Fill: %.1f%%\n",
2982 le64toh(f
->header
->n_fields
),
2983 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))));
2985 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_tags
))
2986 printf("Tag Objects: %"PRIu64
"\n",
2987 le64toh(f
->header
->n_tags
));
2988 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
2989 printf("Entry Array Objects: %"PRIu64
"\n",
2990 le64toh(f
->header
->n_entry_arrays
));
2992 if (fstat(f
->fd
, &st
) >= 0)
2993 printf("Disk usage: %s\n", format_bytes(bytes
, sizeof(bytes
), (uint64_t) st
.st_blocks
* 512ULL));
2996 static int journal_file_warn_btrfs(JournalFile
*f
) {
3002 /* Before we write anything, check if the COW logic is turned
3003 * off on btrfs. Given our write pattern that is quite
3004 * unfriendly to COW file systems this should greatly improve
3005 * performance on COW file systems, such as btrfs, at the
3006 * expense of data integrity features (which shouldn't be too
3007 * bad, given that we do our own checksumming). */
3009 r
= btrfs_is_filesystem(f
->fd
);
3011 return log_warning_errno(r
, "Failed to determine if journal is on btrfs: %m");
3015 r
= read_attr_fd(f
->fd
, &attrs
);
3017 return log_warning_errno(r
, "Failed to read file attributes: %m");
3019 if (attrs
& FS_NOCOW_FL
) {
3020 log_debug("Detected btrfs file system with copy-on-write disabled, all is good.");
3024 log_notice("Creating journal file %s on a btrfs file system, and copy-on-write is enabled. "
3025 "This is likely to slow down journal access substantially, please consider turning "
3026 "off the copy-on-write file attribute on the journal directory, using chattr +C.", f
->path
);
3031 int journal_file_open(
3038 JournalMetrics
*metrics
,
3039 MMapCache
*mmap_cache
,
3040 Set
*deferred_closes
,
3041 JournalFile
*template,
3042 JournalFile
**ret
) {
3044 bool newly_created
= false;
3050 assert(fd
>= 0 || fname
);
3052 if ((flags
& O_ACCMODE
) != O_RDONLY
&&
3053 (flags
& O_ACCMODE
) != O_RDWR
)
3057 if (!endswith(fname
, ".journal") &&
3058 !endswith(fname
, ".journal~"))
3062 f
= new0(JournalFile
, 1);
3070 f
->prot
= prot_from_flags(flags
);
3071 f
->writable
= (flags
& O_ACCMODE
) != O_RDONLY
;
3072 #if defined(HAVE_LZ4)
3073 f
->compress_lz4
= compress
;
3074 #elif defined(HAVE_XZ)
3075 f
->compress_xz
= compress
;
3082 f
->mmap
= mmap_cache_ref(mmap_cache
);
3084 f
->mmap
= mmap_cache_new();
3092 f
->path
= strdup(fname
);
3093 else /* If we don't know the path, fill in something explanatory and vaguely useful */
3094 asprintf(&f
->path
, "/proc/self/%i", fd
);
3100 f
->chain_cache
= ordered_hashmap_new(&uint64_hash_ops
);
3101 if (!f
->chain_cache
) {
3107 f
->fd
= open(f
->path
, f
->flags
|O_CLOEXEC
, f
->mode
);
3113 /* fds we opened here by us should also be closed by us. */
3117 r
= journal_file_fstat(f
);
3121 if (f
->last_stat
.st_size
== 0 && f
->writable
) {
3123 (void) journal_file_warn_btrfs(f
);
3125 /* Let's attach the creation time to the journal file,
3126 * so that the vacuuming code knows the age of this
3127 * file even if the file might end up corrupted one
3128 * day... Ideally we'd just use the creation time many
3129 * file systems maintain for each file, but there is
3130 * currently no usable API to query this, hence let's
3131 * emulate this via extended attributes. If extended
3132 * attributes are not supported we'll just skip this,
3133 * and rely solely on mtime/atime/ctime of the file. */
3135 fd_setcrtime(f
->fd
, 0);
3138 /* Try to load the FSPRG state, and if we can't, then
3139 * just don't do sealing */
3141 r
= journal_file_fss_load(f
);
3147 r
= journal_file_init_header(f
, template);
3151 r
= journal_file_fstat(f
);
3155 newly_created
= true;
3158 if (f
->last_stat
.st_size
< (off_t
) HEADER_SIZE_MIN
) {
3163 r
= mmap_cache_get(f
->mmap
, f
->fd
, f
->prot
, CONTEXT_HEADER
, true, 0, PAGE_ALIGN(sizeof(Header
)), &f
->last_stat
, &h
);
3169 if (!newly_created
) {
3170 if (deferred_closes
)
3171 journal_file_close_set(deferred_closes
);
3173 r
= journal_file_verify_header(f
);
3179 if (!newly_created
&& f
->writable
) {
3180 r
= journal_file_fss_load(f
);
3188 journal_default_metrics(metrics
, f
->fd
);
3189 f
->metrics
= *metrics
;
3190 } else if (template)
3191 f
->metrics
= template->metrics
;
3193 r
= journal_file_refresh_header(f
);
3199 r
= journal_file_hmac_setup(f
);
3204 if (newly_created
) {
3205 r
= journal_file_setup_field_hash_table(f
);
3209 r
= journal_file_setup_data_hash_table(f
);
3214 r
= journal_file_append_first_tag(f
);
3220 if (mmap_cache_got_sigbus(f
->mmap
, f
->fd
)) {
3225 if (template && template->post_change_timer
) {
3226 r
= journal_file_enable_post_change_timer(
3228 sd_event_source_get_event(template->post_change_timer
),
3229 template->post_change_timer_period
);
3235 /* The file is opened now successfully, thus we take possession of any passed in fd. */
3242 if (f
->fd
>= 0 && mmap_cache_got_sigbus(f
->mmap
, f
->fd
))
3245 (void) journal_file_close(f
);
3250 int journal_file_rotate(JournalFile
**f
, bool compress
, bool seal
, Set
*deferred_closes
) {
3251 _cleanup_free_
char *p
= NULL
;
3253 JournalFile
*old_file
, *new_file
= NULL
;
3261 if (!old_file
->writable
)
3264 /* Is this a journal file that was passed to us as fd? If so, we synthesized a path name for it, and we refuse
3265 * rotation, since we don't know the actual path, and couldn't rename the file hence.*/
3266 if (path_startswith(old_file
->path
, "/proc/self/fd"))
3269 if (!endswith(old_file
->path
, ".journal"))
3272 l
= strlen(old_file
->path
);
3273 r
= asprintf(&p
, "%.*s@" SD_ID128_FORMAT_STR
"-%016"PRIx64
"-%016"PRIx64
".journal",
3274 (int) l
- 8, old_file
->path
,
3275 SD_ID128_FORMAT_VAL(old_file
->header
->seqnum_id
),
3276 le64toh((*f
)->header
->head_entry_seqnum
),
3277 le64toh((*f
)->header
->head_entry_realtime
));
3281 /* Try to rename the file to the archived version. If the file
3282 * already was deleted, we'll get ENOENT, let's ignore that
3284 r
= rename(old_file
->path
, p
);
3285 if (r
< 0 && errno
!= ENOENT
)
3288 /* Sync the rename to disk */
3289 (void) fsync_directory_of_file(old_file
->fd
);
3291 /* Set as archive so offlining commits w/state=STATE_ARCHIVED.
3292 * Previously we would set old_file->header->state to STATE_ARCHIVED directly here,
3293 * but journal_file_set_offline() short-circuits when state != STATE_ONLINE, which
3294 * would result in the rotated journal never getting fsync() called before closing.
3295 * Now we simply queue the archive state by setting an archive bit, leaving the state
3296 * as STATE_ONLINE so proper offlining occurs. */
3297 old_file
->archive
= true;
3299 /* Currently, btrfs is not very good with out write patterns
3300 * and fragments heavily. Let's defrag our journal files when
3301 * we archive them */
3302 old_file
->defrag_on_close
= true;
3304 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
);
3306 if (deferred_closes
&&
3307 set_put(deferred_closes
, old_file
) >= 0)
3308 (void) journal_file_set_offline(old_file
, false);
3310 (void) journal_file_close(old_file
);
3316 int journal_file_open_reliably(
3322 JournalMetrics
*metrics
,
3323 MMapCache
*mmap_cache
,
3324 Set
*deferred_closes
,
3325 JournalFile
*template,
3326 JournalFile
**ret
) {
3330 _cleanup_free_
char *p
= NULL
;
3332 r
= journal_file_open(-1, fname
, flags
, mode
, compress
, seal
, metrics
, mmap_cache
, deferred_closes
, template, ret
);
3334 -EBADMSG
, /* corrupted */
3335 -ENODATA
, /* truncated */
3336 -EHOSTDOWN
, /* other machine */
3337 -EPROTONOSUPPORT
, /* incompatible feature */
3338 -EBUSY
, /* unclean shutdown */
3339 -ESHUTDOWN
, /* already archived */
3340 -EIO
, /* IO error, including SIGBUS on mmap */
3341 -EIDRM
, /* File has been deleted */
3342 -ETXTBSY
)) /* File is from the future */
3345 if ((flags
& O_ACCMODE
) == O_RDONLY
)
3348 if (!(flags
& O_CREAT
))
3351 if (!endswith(fname
, ".journal"))
3354 /* The file is corrupted. Rotate it away and try it again (but only once) */
3357 if (asprintf(&p
, "%.*s@%016"PRIx64
"-%016"PRIx64
".journal~",
3359 now(CLOCK_REALTIME
),
3363 if (rename(fname
, p
) < 0)
3366 /* btrfs doesn't cope well with our write pattern and
3367 * fragments heavily. Let's defrag all files we rotate */
3369 (void) chattr_path(p
, 0, FS_NOCOW_FL
);
3370 (void) btrfs_defrag(p
);
3372 log_warning_errno(r
, "File %s corrupted or uncleanly shut down, renaming and replacing.", fname
);
3374 return journal_file_open(-1, fname
, flags
, mode
, compress
, seal
, metrics
, mmap_cache
, deferred_closes
, template, ret
);
3377 int journal_file_copy_entry(JournalFile
*from
, JournalFile
*to
, Object
*o
, uint64_t p
, uint64_t *seqnum
, Object
**ret
, uint64_t *offset
) {
3379 uint64_t q
, xor_hash
= 0;
3392 ts
.monotonic
= le64toh(o
->entry
.monotonic
);
3393 ts
.realtime
= le64toh(o
->entry
.realtime
);
3395 n
= journal_file_entry_n_items(o
);
3396 /* alloca() can't take 0, hence let's allocate at least one */
3397 items
= alloca(sizeof(EntryItem
) * MAX(1u, n
));
3399 for (i
= 0; i
< n
; i
++) {
3406 q
= le64toh(o
->entry
.items
[i
].object_offset
);
3407 le_hash
= o
->entry
.items
[i
].hash
;
3409 r
= journal_file_move_to_object(from
, OBJECT_DATA
, q
, &o
);
3413 if (le_hash
!= o
->data
.hash
)
3416 l
= le64toh(o
->object
.size
) - offsetof(Object
, data
.payload
);
3419 /* We hit the limit on 32bit machines */
3420 if ((uint64_t) t
!= l
)
3423 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
) {
3424 #if defined(HAVE_XZ) || defined(HAVE_LZ4)
3427 r
= decompress_blob(o
->object
.flags
& OBJECT_COMPRESSION_MASK
,
3428 o
->data
.payload
, l
, &from
->compress_buffer
, &from
->compress_buffer_size
, &rsize
, 0);
3432 data
= from
->compress_buffer
;
3435 return -EPROTONOSUPPORT
;
3438 data
= o
->data
.payload
;
3440 r
= journal_file_append_data(to
, data
, l
, &u
, &h
);
3444 xor_hash
^= le64toh(u
->data
.hash
);
3445 items
[i
].object_offset
= htole64(h
);
3446 items
[i
].hash
= u
->data
.hash
;
3448 r
= journal_file_move_to_object(from
, OBJECT_ENTRY
, p
, &o
);
3453 r
= journal_file_append_entry_internal(to
, &ts
, xor_hash
, items
, n
, seqnum
, ret
, offset
);
3455 if (mmap_cache_got_sigbus(to
->mmap
, to
->fd
))
3461 void journal_reset_metrics(JournalMetrics
*m
) {
3464 /* Set everything to "pick automatic values". */
3466 *m
= (JournalMetrics
) {
3467 .min_use
= (uint64_t) -1,
3468 .max_use
= (uint64_t) -1,
3469 .min_size
= (uint64_t) -1,
3470 .max_size
= (uint64_t) -1,
3471 .keep_free
= (uint64_t) -1,
3472 .n_max_files
= (uint64_t) -1,
3476 void journal_default_metrics(JournalMetrics
*m
, int fd
) {
3477 char a
[FORMAT_BYTES_MAX
], b
[FORMAT_BYTES_MAX
], c
[FORMAT_BYTES_MAX
], d
[FORMAT_BYTES_MAX
], e
[FORMAT_BYTES_MAX
];
3484 if (fstatvfs(fd
, &ss
) >= 0)
3485 fs_size
= ss
.f_frsize
* ss
.f_blocks
;
3487 log_debug_errno(errno
, "Failed to detremine disk size: %m");
3491 if (m
->max_use
== (uint64_t) -1) {
3494 m
->max_use
= PAGE_ALIGN(fs_size
/ 10); /* 10% of file system size */
3496 if (m
->max_use
> DEFAULT_MAX_USE_UPPER
)
3497 m
->max_use
= DEFAULT_MAX_USE_UPPER
;
3499 if (m
->max_use
< DEFAULT_MAX_USE_LOWER
)
3500 m
->max_use
= DEFAULT_MAX_USE_LOWER
;
3502 m
->max_use
= DEFAULT_MAX_USE_LOWER
;
3504 m
->max_use
= PAGE_ALIGN(m
->max_use
);
3506 if (m
->max_use
!= 0 && m
->max_use
< JOURNAL_FILE_SIZE_MIN
*2)
3507 m
->max_use
= JOURNAL_FILE_SIZE_MIN
*2;
3510 if (m
->min_use
== (uint64_t) -1)
3511 m
->min_use
= DEFAULT_MIN_USE
;
3513 if (m
->min_use
> m
->max_use
)
3514 m
->min_use
= m
->max_use
;
3516 if (m
->max_size
== (uint64_t) -1) {
3517 m
->max_size
= PAGE_ALIGN(m
->max_use
/ 8); /* 8 chunks */
3519 if (m
->max_size
> DEFAULT_MAX_SIZE_UPPER
)
3520 m
->max_size
= DEFAULT_MAX_SIZE_UPPER
;
3522 m
->max_size
= PAGE_ALIGN(m
->max_size
);
3524 if (m
->max_size
!= 0) {
3525 if (m
->max_size
< JOURNAL_FILE_SIZE_MIN
)
3526 m
->max_size
= JOURNAL_FILE_SIZE_MIN
;
3528 if (m
->max_use
!= 0 && m
->max_size
*2 > m
->max_use
)
3529 m
->max_use
= m
->max_size
*2;
3532 if (m
->min_size
== (uint64_t) -1)
3533 m
->min_size
= JOURNAL_FILE_SIZE_MIN
;
3535 m
->min_size
= PAGE_ALIGN(m
->min_size
);
3537 if (m
->min_size
< JOURNAL_FILE_SIZE_MIN
)
3538 m
->min_size
= JOURNAL_FILE_SIZE_MIN
;
3540 if (m
->max_size
!= 0 && m
->min_size
> m
->max_size
)
3541 m
->max_size
= m
->min_size
;
3544 if (m
->keep_free
== (uint64_t) -1) {
3547 m
->keep_free
= PAGE_ALIGN(fs_size
* 3 / 20); /* 15% of file system size */
3549 if (m
->keep_free
> DEFAULT_KEEP_FREE_UPPER
)
3550 m
->keep_free
= DEFAULT_KEEP_FREE_UPPER
;
3553 m
->keep_free
= DEFAULT_KEEP_FREE
;
3556 if (m
->n_max_files
== (uint64_t) -1)
3557 m
->n_max_files
= DEFAULT_N_MAX_FILES
;
3559 log_debug("Fixed min_use=%s max_use=%s max_size=%s min_size=%s keep_free=%s n_max_files=%" PRIu64
,
3560 format_bytes(a
, sizeof(a
), m
->min_use
),
3561 format_bytes(b
, sizeof(b
), m
->max_use
),
3562 format_bytes(c
, sizeof(c
), m
->max_size
),
3563 format_bytes(d
, sizeof(d
), m
->min_size
),
3564 format_bytes(e
, sizeof(e
), m
->keep_free
),
3568 int journal_file_get_cutoff_realtime_usec(JournalFile
*f
, usec_t
*from
, usec_t
*to
) {
3574 if (f
->header
->head_entry_realtime
== 0)
3577 *from
= le64toh(f
->header
->head_entry_realtime
);
3581 if (f
->header
->tail_entry_realtime
== 0)
3584 *to
= le64toh(f
->header
->tail_entry_realtime
);
3590 int journal_file_get_cutoff_monotonic_usec(JournalFile
*f
, sd_id128_t boot_id
, usec_t
*from
, usec_t
*to
) {
3598 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, &p
);
3602 if (le64toh(o
->data
.n_entries
) <= 0)
3606 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, le64toh(o
->data
.entry_offset
), &o
);
3610 *from
= le64toh(o
->entry
.monotonic
);
3614 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
3618 r
= generic_array_get_plus_one(f
,
3619 le64toh(o
->data
.entry_offset
),
3620 le64toh(o
->data
.entry_array_offset
),
3621 le64toh(o
->data
.n_entries
)-1,
3626 *to
= le64toh(o
->entry
.monotonic
);
3632 bool journal_file_rotate_suggested(JournalFile
*f
, usec_t max_file_usec
) {
3636 /* If we gained new header fields we gained new features,
3637 * hence suggest a rotation */
3638 if (le64toh(f
->header
->header_size
) < sizeof(Header
)) {
3639 log_debug("%s uses an outdated header, suggesting rotation.", f
->path
);
3643 /* Let's check if the hash tables grew over a certain fill
3644 * level (75%, borrowing this value from Java's hash table
3645 * implementation), and if so suggest a rotation. To calculate
3646 * the fill level we need the n_data field, which only exists
3647 * in newer versions. */
3649 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
3650 if (le64toh(f
->header
->n_data
) * 4ULL > (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
3651 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.",
3653 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))),
3654 le64toh(f
->header
->n_data
),
3655 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
3656 (unsigned long long) f
->last_stat
.st_size
,
3657 f
->last_stat
.st_size
/ le64toh(f
->header
->n_data
));
3661 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
3662 if (le64toh(f
->header
->n_fields
) * 4ULL > (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
3663 log_debug("Field hash table of %s has a fill level at %.1f (%"PRIu64
" of %"PRIu64
" items), suggesting rotation.",
3665 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))),
3666 le64toh(f
->header
->n_fields
),
3667 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
));
3671 /* Are the data objects properly indexed by field objects? */
3672 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
) &&
3673 JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
) &&
3674 le64toh(f
->header
->n_data
) > 0 &&
3675 le64toh(f
->header
->n_fields
) == 0)
3678 if (max_file_usec
> 0) {
3681 h
= le64toh(f
->header
->head_entry_realtime
);
3682 t
= now(CLOCK_REALTIME
);
3684 if (h
> 0 && t
> h
+ max_file_usec
)