2 This file is part of systemd.
4 Copyright 2011 Lennart Poettering
6 systemd is free software; you can redistribute it and/or modify it
7 under the terms of the GNU Lesser General Public License as published by
8 the Free Software Foundation; either version 2.1 of the License, or
9 (at your option) any later version.
11 systemd is distributed in the hope that it will be useful, but
12 WITHOUT ANY WARRANTY; without even the implied warranty of
13 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 Lesser General Public License for more details.
16 You should have received a copy of the GNU Lesser General Public License
17 along with systemd; If not, see <http://www.gnu.org/licenses/>.
26 #include <sys/statvfs.h>
30 #include "alloc-util.h"
31 #include "btrfs-util.h"
32 #include "chattr-util.h"
35 #include "journal-authenticate.h"
36 #include "journal-def.h"
37 #include "journal-file.h"
39 #include "parse-util.h"
40 #include "path-util.h"
41 #include "random-util.h"
44 #include "string-util.h"
46 #include "xattr-util.h"
48 #define DEFAULT_DATA_HASH_TABLE_SIZE (2047ULL*sizeof(HashItem))
49 #define DEFAULT_FIELD_HASH_TABLE_SIZE (333ULL*sizeof(HashItem))
51 #define COMPRESSION_SIZE_THRESHOLD (512ULL)
53 /* This is the minimum journal file size */
54 #define JOURNAL_FILE_SIZE_MIN (512ULL*1024ULL) /* 512 KiB */
56 /* These are the lower and upper bounds if we deduce the max_use value
57 * from the file system size */
58 #define DEFAULT_MAX_USE_LOWER (1ULL*1024ULL*1024ULL) /* 1 MiB */
59 #define DEFAULT_MAX_USE_UPPER (4ULL*1024ULL*1024ULL*1024ULL) /* 4 GiB */
61 /* This is the default minimal use limit, how much we'll use even if keep_free suggests otherwise. */
62 #define DEFAULT_MIN_USE (1ULL*1024ULL*1024ULL) /* 1 MiB */
64 /* This is the upper bound if we deduce max_size from max_use */
65 #define DEFAULT_MAX_SIZE_UPPER (128ULL*1024ULL*1024ULL) /* 128 MiB */
67 /* This is the upper bound if we deduce the keep_free value from the
69 #define DEFAULT_KEEP_FREE_UPPER (4ULL*1024ULL*1024ULL*1024ULL) /* 4 GiB */
71 /* This is the keep_free value when we can't determine the system
73 #define DEFAULT_KEEP_FREE (1024ULL*1024ULL) /* 1 MB */
75 /* This is the default maximum number of journal files to keep around. */
76 #define DEFAULT_N_MAX_FILES (100)
78 /* n_data was the first entry we added after the initial file format design */
79 #define HEADER_SIZE_MIN ALIGN64(offsetof(Header, n_data))
81 /* How many entries to keep in the entry array chain cache at max */
82 #define CHAIN_CACHE_MAX 20
84 /* How much to increase the journal file size at once each time we allocate something new. */
85 #define FILE_SIZE_INCREASE (8ULL*1024ULL*1024ULL) /* 8MB */
87 /* Reread fstat() of the file for detecting deletions at least this often */
88 #define LAST_STAT_REFRESH_USEC (5*USEC_PER_SEC)
90 /* The mmap context to use for the header we pick as one above the last defined typed */
91 #define CONTEXT_HEADER _OBJECT_TYPE_MAX
93 /* This may be called from a separate thread to prevent blocking the caller for the duration of fsync().
94 * As a result we use atomic operations on f->offline_state for inter-thread communications with
95 * journal_file_set_offline() and journal_file_set_online(). */
96 static void journal_file_set_offline_internal(JournalFile
*f
) {
102 switch (f
->offline_state
) {
104 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_CANCEL
, OFFLINE_DONE
))
108 case OFFLINE_AGAIN_FROM_SYNCING
:
109 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_AGAIN_FROM_SYNCING
, OFFLINE_SYNCING
))
113 case OFFLINE_AGAIN_FROM_OFFLINING
:
114 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_AGAIN_FROM_OFFLINING
, OFFLINE_SYNCING
))
118 case OFFLINE_SYNCING
:
121 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_SYNCING
, OFFLINE_OFFLINING
))
124 f
->header
->state
= f
->archive
? STATE_ARCHIVED
: STATE_OFFLINE
;
128 case OFFLINE_OFFLINING
:
129 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_OFFLINING
, OFFLINE_DONE
))
137 log_debug("OFFLINE_JOINED unexpected offline state for journal_file_set_offline_internal()");
143 static void * journal_file_set_offline_thread(void *arg
) {
144 JournalFile
*f
= arg
;
146 journal_file_set_offline_internal(f
);
151 static int journal_file_set_offline_thread_join(JournalFile
*f
) {
156 if (f
->offline_state
== OFFLINE_JOINED
)
159 r
= pthread_join(f
->offline_thread
, NULL
);
163 f
->offline_state
= OFFLINE_JOINED
;
165 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
171 /* Trigger a restart if the offline thread is mid-flight in a restartable state. */
172 static bool journal_file_set_offline_try_restart(JournalFile
*f
) {
174 switch (f
->offline_state
) {
175 case OFFLINE_AGAIN_FROM_SYNCING
:
176 case OFFLINE_AGAIN_FROM_OFFLINING
:
180 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_CANCEL
, OFFLINE_AGAIN_FROM_SYNCING
))
184 case OFFLINE_SYNCING
:
185 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_SYNCING
, OFFLINE_AGAIN_FROM_SYNCING
))
189 case OFFLINE_OFFLINING
:
190 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_OFFLINING
, OFFLINE_AGAIN_FROM_OFFLINING
))
200 /* Sets a journal offline.
202 * If wait is false then an offline is dispatched in a separate thread for a
203 * subsequent journal_file_set_offline() or journal_file_set_online() of the
204 * same journal to synchronize with.
206 * If wait is true, then either an existing offline thread will be restarted
207 * and joined, or if none exists the offline is simply performed in this
208 * context without involving another thread.
210 int journal_file_set_offline(JournalFile
*f
, bool wait
) {
219 if (!(f
->fd
>= 0 && f
->header
))
222 /* An offlining journal is implicitly online and may modify f->header->state,
223 * we must also join any potentially lingering offline thread when not online. */
224 if (!journal_file_is_offlining(f
) && f
->header
->state
!= STATE_ONLINE
)
225 return journal_file_set_offline_thread_join(f
);
227 /* Restart an in-flight offline thread and wait if needed, or join a lingering done one. */
228 restarted
= journal_file_set_offline_try_restart(f
);
229 if ((restarted
&& wait
) || !restarted
) {
230 r
= journal_file_set_offline_thread_join(f
);
238 /* Initiate a new offline. */
239 f
->offline_state
= OFFLINE_SYNCING
;
241 if (wait
) /* Without using a thread if waiting. */
242 journal_file_set_offline_internal(f
);
244 r
= pthread_create(&f
->offline_thread
, NULL
, journal_file_set_offline_thread
, f
);
246 f
->offline_state
= OFFLINE_JOINED
;
254 static int journal_file_set_online(JournalFile
*f
) {
262 if (!(f
->fd
>= 0 && f
->header
))
266 switch (f
->offline_state
) {
268 /* No offline thread, no need to wait. */
272 case OFFLINE_SYNCING
:
273 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_SYNCING
, OFFLINE_CANCEL
))
275 /* Canceled syncing prior to offlining, no need to wait. */
278 case OFFLINE_AGAIN_FROM_SYNCING
:
279 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_AGAIN_FROM_SYNCING
, OFFLINE_CANCEL
))
281 /* Canceled restart from syncing, no need to wait. */
284 case OFFLINE_AGAIN_FROM_OFFLINING
:
285 if (!__sync_bool_compare_and_swap(&f
->offline_state
, OFFLINE_AGAIN_FROM_OFFLINING
, OFFLINE_CANCEL
))
287 /* Canceled restart from offlining, must wait for offlining to complete however. */
293 r
= journal_file_set_offline_thread_join(f
);
303 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
306 switch (f
->header
->state
) {
311 f
->header
->state
= STATE_ONLINE
;
320 bool journal_file_is_offlining(JournalFile
*f
) {
323 __sync_synchronize();
325 if (IN_SET(f
->offline_state
, OFFLINE_DONE
, 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
->cache_fd
)
359 mmap_cache_free_fd(f
->mmap
, f
->cache_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 HAVE_XZ || 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
);
400 void journal_file_close_set(Set
*s
) {
405 while ((f
= set_steal_first(s
)))
406 (void) journal_file_close(f
);
409 static int journal_file_init_header(JournalFile
*f
, JournalFile
*template) {
416 memcpy(h
.signature
, HEADER_SIGNATURE
, 8);
417 h
.header_size
= htole64(ALIGN64(sizeof(h
)));
419 h
.incompatible_flags
|= htole32(
420 f
->compress_xz
* HEADER_INCOMPATIBLE_COMPRESSED_XZ
|
421 f
->compress_lz4
* HEADER_INCOMPATIBLE_COMPRESSED_LZ4
);
423 h
.compatible_flags
= htole32(
424 f
->seal
* HEADER_COMPATIBLE_SEALED
);
426 r
= sd_id128_randomize(&h
.file_id
);
431 h
.seqnum_id
= template->header
->seqnum_id
;
432 h
.tail_entry_seqnum
= template->header
->tail_entry_seqnum
;
434 h
.seqnum_id
= h
.file_id
;
436 k
= pwrite(f
->fd
, &h
, sizeof(h
), 0);
446 static int fsync_directory_of_file(int fd
) {
447 _cleanup_free_
char *path
= NULL
, *dn
= NULL
;
448 _cleanup_close_
int dfd
= -1;
452 if (fstat(fd
, &st
) < 0)
455 if (!S_ISREG(st
.st_mode
))
458 r
= fd_get_path(fd
, &path
);
462 if (!path_is_absolute(path
))
465 dn
= dirname_malloc(path
);
469 dfd
= open(dn
, O_RDONLY
|O_CLOEXEC
|O_DIRECTORY
);
479 static int journal_file_refresh_header(JournalFile
*f
) {
486 r
= sd_id128_get_machine(&f
->header
->machine_id
);
490 r
= sd_id128_get_boot(&boot_id
);
494 if (sd_id128_equal(boot_id
, f
->header
->boot_id
))
495 f
->tail_entry_monotonic_valid
= true;
497 f
->header
->boot_id
= boot_id
;
499 r
= journal_file_set_online(f
);
501 /* Sync the online state to disk */
504 /* We likely just created a new file, also sync the directory this file is located in. */
505 (void) fsync_directory_of_file(f
->fd
);
510 static bool warn_wrong_flags(const JournalFile
*f
, bool compatible
) {
511 const uint32_t any
= compatible
? HEADER_COMPATIBLE_ANY
: HEADER_INCOMPATIBLE_ANY
,
512 supported
= compatible
? HEADER_COMPATIBLE_SUPPORTED
: HEADER_INCOMPATIBLE_SUPPORTED
;
513 const char *type
= compatible
? "compatible" : "incompatible";
516 flags
= le32toh(compatible
? f
->header
->compatible_flags
: f
->header
->incompatible_flags
);
518 if (flags
& ~supported
) {
520 log_debug("Journal file %s has unknown %s flags 0x%"PRIx32
,
521 f
->path
, type
, flags
& ~any
);
522 flags
= (flags
& any
) & ~supported
;
526 _cleanup_free_
char *t
= NULL
;
528 if (compatible
&& (flags
& HEADER_COMPATIBLE_SEALED
))
529 strv
[n
++] = "sealed";
530 if (!compatible
&& (flags
& HEADER_INCOMPATIBLE_COMPRESSED_XZ
))
531 strv
[n
++] = "xz-compressed";
532 if (!compatible
&& (flags
& HEADER_INCOMPATIBLE_COMPRESSED_LZ4
))
533 strv
[n
++] = "lz4-compressed";
535 assert(n
< ELEMENTSOF(strv
));
537 t
= strv_join((char**) strv
, ", ");
538 log_debug("Journal file %s uses %s %s %s disabled at compilation time.",
539 f
->path
, type
, n
> 1 ? "flags" : "flag", strnull(t
));
547 static int journal_file_verify_header(JournalFile
*f
) {
548 uint64_t arena_size
, header_size
;
553 if (memcmp(f
->header
->signature
, HEADER_SIGNATURE
, 8))
556 /* In both read and write mode we refuse to open files with incompatible
557 * flags we don't know. */
558 if (warn_wrong_flags(f
, false))
559 return -EPROTONOSUPPORT
;
561 /* When open for writing we refuse to open files with compatible flags, too. */
562 if (f
->writable
&& warn_wrong_flags(f
, true))
563 return -EPROTONOSUPPORT
;
565 if (f
->header
->state
>= _STATE_MAX
)
568 header_size
= le64toh(f
->header
->header_size
);
570 /* The first addition was n_data, so check that we are at least this large */
571 if (header_size
< HEADER_SIZE_MIN
)
574 if (JOURNAL_HEADER_SEALED(f
->header
) && !JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
577 arena_size
= le64toh(f
->header
->arena_size
);
579 if (UINT64_MAX
- header_size
< arena_size
|| header_size
+ arena_size
> (uint64_t) f
->last_stat
.st_size
)
582 if (le64toh(f
->header
->tail_object_offset
) > header_size
+ arena_size
)
585 if (!VALID64(le64toh(f
->header
->data_hash_table_offset
)) ||
586 !VALID64(le64toh(f
->header
->field_hash_table_offset
)) ||
587 !VALID64(le64toh(f
->header
->tail_object_offset
)) ||
588 !VALID64(le64toh(f
->header
->entry_array_offset
)))
592 sd_id128_t machine_id
;
596 r
= sd_id128_get_machine(&machine_id
);
600 if (!sd_id128_equal(machine_id
, f
->header
->machine_id
))
603 state
= f
->header
->state
;
605 if (state
== STATE_ARCHIVED
)
606 return -ESHUTDOWN
; /* Already archived */
607 else if (state
== STATE_ONLINE
) {
608 log_debug("Journal file %s is already online. Assuming unclean closing.", f
->path
);
610 } else if (state
!= STATE_OFFLINE
) {
611 log_debug("Journal file %s has unknown state %i.", f
->path
, state
);
615 if (f
->header
->field_hash_table_size
== 0 || f
->header
->data_hash_table_size
== 0)
618 /* Don't permit appending to files from the future. Because otherwise the realtime timestamps wouldn't
619 * be strictly ordered in the entries in the file anymore, and we can't have that since it breaks
621 if (le64toh(f
->header
->tail_entry_realtime
) > now(CLOCK_REALTIME
)) {
622 log_debug("Journal file %s is from the future, refusing to append new data to it that'd be older.", f
->path
);
627 f
->compress_xz
= JOURNAL_HEADER_COMPRESSED_XZ(f
->header
);
628 f
->compress_lz4
= JOURNAL_HEADER_COMPRESSED_LZ4(f
->header
);
630 f
->seal
= JOURNAL_HEADER_SEALED(f
->header
);
635 static int journal_file_fstat(JournalFile
*f
) {
639 if (fstat(f
->fd
, &f
->last_stat
) < 0)
642 f
->last_stat_usec
= now(CLOCK_MONOTONIC
);
644 /* Refuse appending to files that are already deleted */
645 if (f
->last_stat
.st_nlink
<= 0)
651 static int journal_file_allocate(JournalFile
*f
, uint64_t offset
, uint64_t size
) {
652 uint64_t old_size
, new_size
;
658 /* We assume that this file is not sparse, and we know that
659 * for sure, since we always call posix_fallocate()
662 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
666 le64toh(f
->header
->header_size
) +
667 le64toh(f
->header
->arena_size
);
669 new_size
= PAGE_ALIGN(offset
+ size
);
670 if (new_size
< le64toh(f
->header
->header_size
))
671 new_size
= le64toh(f
->header
->header_size
);
673 if (new_size
<= old_size
) {
675 /* We already pre-allocated enough space, but before
676 * we write to it, let's check with fstat() if the
677 * file got deleted, in order make sure we don't throw
678 * away the data immediately. Don't check fstat() for
679 * all writes though, but only once ever 10s. */
681 if (f
->last_stat_usec
+ LAST_STAT_REFRESH_USEC
> now(CLOCK_MONOTONIC
))
684 return journal_file_fstat(f
);
687 /* Allocate more space. */
689 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
692 if (new_size
> f
->metrics
.min_size
&& f
->metrics
.keep_free
> 0) {
695 if (fstatvfs(f
->fd
, &svfs
) >= 0) {
698 available
= LESS_BY((uint64_t) svfs
.f_bfree
* (uint64_t) svfs
.f_bsize
, f
->metrics
.keep_free
);
700 if (new_size
- old_size
> available
)
705 /* Increase by larger blocks at once */
706 new_size
= ((new_size
+FILE_SIZE_INCREASE
-1) / FILE_SIZE_INCREASE
) * FILE_SIZE_INCREASE
;
707 if (f
->metrics
.max_size
> 0 && new_size
> f
->metrics
.max_size
)
708 new_size
= f
->metrics
.max_size
;
710 /* Note that the glibc fallocate() fallback is very
711 inefficient, hence we try to minimize the allocation area
713 r
= posix_fallocate(f
->fd
, old_size
, new_size
- old_size
);
717 f
->header
->arena_size
= htole64(new_size
- le64toh(f
->header
->header_size
));
719 return journal_file_fstat(f
);
722 static unsigned type_to_context(ObjectType type
) {
723 /* One context for each type, plus one catch-all for the rest */
724 assert_cc(_OBJECT_TYPE_MAX
<= MMAP_CACHE_MAX_CONTEXTS
);
725 assert_cc(CONTEXT_HEADER
< MMAP_CACHE_MAX_CONTEXTS
);
726 return type
> OBJECT_UNUSED
&& type
< _OBJECT_TYPE_MAX
? type
: 0;
729 static int journal_file_move_to(JournalFile
*f
, ObjectType type
, bool keep_always
, uint64_t offset
, uint64_t size
, void **ret
, size_t *ret_size
) {
738 /* Avoid SIGBUS on invalid accesses */
739 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
) {
740 /* Hmm, out of range? Let's refresh the fstat() data
741 * first, before we trust that check. */
743 r
= journal_file_fstat(f
);
747 if (offset
+ size
> (uint64_t) f
->last_stat
.st_size
)
748 return -EADDRNOTAVAIL
;
751 return mmap_cache_get(f
->mmap
, f
->cache_fd
, f
->prot
, type_to_context(type
), keep_always
, offset
, size
, &f
->last_stat
, ret
, ret_size
);
754 static uint64_t minimum_header_size(Object
*o
) {
756 static const uint64_t table
[] = {
757 [OBJECT_DATA
] = sizeof(DataObject
),
758 [OBJECT_FIELD
] = sizeof(FieldObject
),
759 [OBJECT_ENTRY
] = sizeof(EntryObject
),
760 [OBJECT_DATA_HASH_TABLE
] = sizeof(HashTableObject
),
761 [OBJECT_FIELD_HASH_TABLE
] = sizeof(HashTableObject
),
762 [OBJECT_ENTRY_ARRAY
] = sizeof(EntryArrayObject
),
763 [OBJECT_TAG
] = sizeof(TagObject
),
766 if (o
->object
.type
>= ELEMENTSOF(table
) || table
[o
->object
.type
] <= 0)
767 return sizeof(ObjectHeader
);
769 return table
[o
->object
.type
];
772 /* Lightweight object checks. We want this to be fast, so that we won't
773 * slowdown every journal_file_move_to_object() call too much. */
774 static int journal_file_check_object(JournalFile
*f
, uint64_t offset
, Object
*o
) {
778 switch (o
->object
.type
) {
781 if ((le64toh(o
->data
.entry_offset
) == 0) ^ (le64toh(o
->data
.n_entries
) == 0)) {
782 log_debug("Bad n_entries: %"PRIu64
": %"PRIu64
,
783 le64toh(o
->data
.n_entries
), offset
);
787 if (le64toh(o
->object
.size
) - offsetof(DataObject
, payload
) <= 0) {
788 log_debug("Bad object size (<= %zu): %"PRIu64
": %"PRIu64
,
789 offsetof(DataObject
, payload
),
790 le64toh(o
->object
.size
),
795 if (!VALID64(le64toh(o
->data
.next_hash_offset
)) ||
796 !VALID64(le64toh(o
->data
.next_field_offset
)) ||
797 !VALID64(le64toh(o
->data
.entry_offset
)) ||
798 !VALID64(le64toh(o
->data
.entry_array_offset
))) {
799 log_debug("Invalid offset, next_hash_offset="OFSfmt
", next_field_offset="OFSfmt
800 ", entry_offset="OFSfmt
", entry_array_offset="OFSfmt
": %"PRIu64
,
801 le64toh(o
->data
.next_hash_offset
),
802 le64toh(o
->data
.next_field_offset
),
803 le64toh(o
->data
.entry_offset
),
804 le64toh(o
->data
.entry_array_offset
),
813 if (le64toh(o
->object
.size
) - offsetof(FieldObject
, payload
) <= 0) {
815 "Bad field size (<= %zu): %"PRIu64
": %"PRIu64
,
816 offsetof(FieldObject
, payload
),
817 le64toh(o
->object
.size
),
822 if (!VALID64(le64toh(o
->field
.next_hash_offset
)) ||
823 !VALID64(le64toh(o
->field
.head_data_offset
))) {
825 "Invalid offset, next_hash_offset="OFSfmt
826 ", head_data_offset="OFSfmt
": %"PRIu64
,
827 le64toh(o
->field
.next_hash_offset
),
828 le64toh(o
->field
.head_data_offset
),
835 if ((le64toh(o
->object
.size
) - offsetof(EntryObject
, items
)) % sizeof(EntryItem
) != 0) {
837 "Bad entry size (<= %zu): %"PRIu64
": %"PRIu64
,
838 offsetof(EntryObject
, items
),
839 le64toh(o
->object
.size
),
844 if ((le64toh(o
->object
.size
) - offsetof(EntryObject
, items
)) / sizeof(EntryItem
) <= 0) {
846 "Invalid number items in entry: %"PRIu64
": %"PRIu64
,
847 (le64toh(o
->object
.size
) - offsetof(EntryObject
, items
)) / sizeof(EntryItem
),
852 if (le64toh(o
->entry
.seqnum
) <= 0) {
854 "Invalid entry seqnum: %"PRIx64
": %"PRIu64
,
855 le64toh(o
->entry
.seqnum
),
860 if (!VALID_REALTIME(le64toh(o
->entry
.realtime
))) {
862 "Invalid entry realtime timestamp: %"PRIu64
": %"PRIu64
,
863 le64toh(o
->entry
.realtime
),
868 if (!VALID_MONOTONIC(le64toh(o
->entry
.monotonic
))) {
870 "Invalid entry monotonic timestamp: %"PRIu64
": %"PRIu64
,
871 le64toh(o
->entry
.monotonic
),
878 case OBJECT_DATA_HASH_TABLE
:
879 case OBJECT_FIELD_HASH_TABLE
:
880 if ((le64toh(o
->object
.size
) - offsetof(HashTableObject
, items
)) % sizeof(HashItem
) != 0 ||
881 (le64toh(o
->object
.size
) - offsetof(HashTableObject
, items
)) / sizeof(HashItem
) <= 0) {
883 "Invalid %s hash table size: %"PRIu64
": %"PRIu64
,
884 o
->object
.type
== OBJECT_DATA_HASH_TABLE
? "data" : "field",
885 le64toh(o
->object
.size
),
892 case OBJECT_ENTRY_ARRAY
:
893 if ((le64toh(o
->object
.size
) - offsetof(EntryArrayObject
, items
)) % sizeof(le64_t
) != 0 ||
894 (le64toh(o
->object
.size
) - offsetof(EntryArrayObject
, items
)) / sizeof(le64_t
) <= 0) {
896 "Invalid object entry array size: %"PRIu64
": %"PRIu64
,
897 le64toh(o
->object
.size
),
902 if (!VALID64(le64toh(o
->entry_array
.next_entry_array_offset
))) {
904 "Invalid object entry array next_entry_array_offset: "OFSfmt
": %"PRIu64
,
905 le64toh(o
->entry_array
.next_entry_array_offset
),
913 if (le64toh(o
->object
.size
) != sizeof(TagObject
)) {
915 "Invalid object tag size: %"PRIu64
": %"PRIu64
,
916 le64toh(o
->object
.size
),
921 if (!VALID_EPOCH(le64toh(o
->tag
.epoch
))) {
923 "Invalid object tag epoch: %"PRIu64
": %"PRIu64
,
924 le64toh(o
->tag
.epoch
),
935 int journal_file_move_to_object(JournalFile
*f
, ObjectType type
, uint64_t offset
, Object
**ret
) {
945 /* Objects may only be located at multiple of 64 bit */
946 if (!VALID64(offset
)) {
947 log_debug("Attempt to move to object at non-64bit boundary: %" PRIu64
, offset
);
951 /* Object may not be located in the file header */
952 if (offset
< le64toh(f
->header
->header_size
)) {
953 log_debug("Attempt to move to object located in file header: %" PRIu64
, offset
);
957 r
= journal_file_move_to(f
, type
, false, offset
, sizeof(ObjectHeader
), &t
, &tsize
);
962 s
= le64toh(o
->object
.size
);
965 log_debug("Attempt to move to uninitialized object: %" PRIu64
, offset
);
968 if (s
< sizeof(ObjectHeader
)) {
969 log_debug("Attempt to move to overly short object: %" PRIu64
, offset
);
973 if (o
->object
.type
<= OBJECT_UNUSED
) {
974 log_debug("Attempt to move to object with invalid type: %" PRIu64
, offset
);
978 if (s
< minimum_header_size(o
)) {
979 log_debug("Attempt to move to truncated object: %" PRIu64
, offset
);
983 if (type
> OBJECT_UNUSED
&& o
->object
.type
!= type
) {
984 log_debug("Attempt to move to object of unexpected type: %" PRIu64
, offset
);
989 r
= journal_file_move_to(f
, type
, false, offset
, s
, &t
, NULL
);
996 r
= journal_file_check_object(f
, offset
, o
);
1004 static uint64_t journal_file_entry_seqnum(JournalFile
*f
, uint64_t *seqnum
) {
1010 r
= le64toh(f
->header
->tail_entry_seqnum
) + 1;
1013 /* If an external seqnum counter was passed, we update
1014 * both the local and the external one, and set it to
1015 * the maximum of both */
1017 if (*seqnum
+ 1 > r
)
1023 f
->header
->tail_entry_seqnum
= htole64(r
);
1025 if (f
->header
->head_entry_seqnum
== 0)
1026 f
->header
->head_entry_seqnum
= htole64(r
);
1031 int journal_file_append_object(JournalFile
*f
, ObjectType type
, uint64_t size
, Object
**ret
, uint64_t *offset
) {
1039 assert(type
> OBJECT_UNUSED
&& type
< _OBJECT_TYPE_MAX
);
1040 assert(size
>= sizeof(ObjectHeader
));
1044 r
= journal_file_set_online(f
);
1048 p
= le64toh(f
->header
->tail_object_offset
);
1050 p
= le64toh(f
->header
->header_size
);
1052 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &tail
);
1056 p
+= ALIGN64(le64toh(tail
->object
.size
));
1059 r
= journal_file_allocate(f
, p
, size
);
1063 r
= journal_file_move_to(f
, type
, false, p
, size
, &t
, NULL
);
1070 o
->object
.type
= type
;
1071 o
->object
.size
= htole64(size
);
1073 f
->header
->tail_object_offset
= htole64(p
);
1074 f
->header
->n_objects
= htole64(le64toh(f
->header
->n_objects
) + 1);
1082 static int journal_file_setup_data_hash_table(JournalFile
*f
) {
1090 /* We estimate that we need 1 hash table entry per 768 bytes
1091 of journal file and we want to make sure we never get
1092 beyond 75% fill level. Calculate the hash table size for
1093 the maximum file size based on these metrics. */
1095 s
= (f
->metrics
.max_size
* 4 / 768 / 3) * sizeof(HashItem
);
1096 if (s
< DEFAULT_DATA_HASH_TABLE_SIZE
)
1097 s
= DEFAULT_DATA_HASH_TABLE_SIZE
;
1099 log_debug("Reserving %"PRIu64
" entries in hash table.", s
/ sizeof(HashItem
));
1101 r
= journal_file_append_object(f
,
1102 OBJECT_DATA_HASH_TABLE
,
1103 offsetof(Object
, hash_table
.items
) + s
,
1108 memzero(o
->hash_table
.items
, s
);
1110 f
->header
->data_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
1111 f
->header
->data_hash_table_size
= htole64(s
);
1116 static int journal_file_setup_field_hash_table(JournalFile
*f
) {
1124 /* We use a fixed size hash table for the fields as this
1125 * number should grow very slowly only */
1127 s
= DEFAULT_FIELD_HASH_TABLE_SIZE
;
1128 r
= journal_file_append_object(f
,
1129 OBJECT_FIELD_HASH_TABLE
,
1130 offsetof(Object
, hash_table
.items
) + s
,
1135 memzero(o
->hash_table
.items
, s
);
1137 f
->header
->field_hash_table_offset
= htole64(p
+ offsetof(Object
, hash_table
.items
));
1138 f
->header
->field_hash_table_size
= htole64(s
);
1143 int journal_file_map_data_hash_table(JournalFile
*f
) {
1151 if (f
->data_hash_table
)
1154 p
= le64toh(f
->header
->data_hash_table_offset
);
1155 s
= le64toh(f
->header
->data_hash_table_size
);
1157 r
= journal_file_move_to(f
,
1158 OBJECT_DATA_HASH_TABLE
,
1165 f
->data_hash_table
= t
;
1169 int journal_file_map_field_hash_table(JournalFile
*f
) {
1177 if (f
->field_hash_table
)
1180 p
= le64toh(f
->header
->field_hash_table_offset
);
1181 s
= le64toh(f
->header
->field_hash_table_size
);
1183 r
= journal_file_move_to(f
,
1184 OBJECT_FIELD_HASH_TABLE
,
1191 f
->field_hash_table
= t
;
1195 static int journal_file_link_field(
1206 assert(f
->field_hash_table
);
1210 if (o
->object
.type
!= OBJECT_FIELD
)
1213 m
= le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
);
1217 /* This might alter the window we are looking at */
1218 o
->field
.next_hash_offset
= o
->field
.head_data_offset
= 0;
1221 p
= le64toh(f
->field_hash_table
[h
].tail_hash_offset
);
1223 f
->field_hash_table
[h
].head_hash_offset
= htole64(offset
);
1225 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1229 o
->field
.next_hash_offset
= htole64(offset
);
1232 f
->field_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1234 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
1235 f
->header
->n_fields
= htole64(le64toh(f
->header
->n_fields
) + 1);
1240 static int journal_file_link_data(
1251 assert(f
->data_hash_table
);
1255 if (o
->object
.type
!= OBJECT_DATA
)
1258 m
= le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
);
1262 /* This might alter the window we are looking at */
1263 o
->data
.next_hash_offset
= o
->data
.next_field_offset
= 0;
1264 o
->data
.entry_offset
= o
->data
.entry_array_offset
= 0;
1265 o
->data
.n_entries
= 0;
1268 p
= le64toh(f
->data_hash_table
[h
].tail_hash_offset
);
1270 /* Only entry in the hash table is easy */
1271 f
->data_hash_table
[h
].head_hash_offset
= htole64(offset
);
1273 /* Move back to the previous data object, to patch in
1276 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1280 o
->data
.next_hash_offset
= htole64(offset
);
1283 f
->data_hash_table
[h
].tail_hash_offset
= htole64(offset
);
1285 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
1286 f
->header
->n_data
= htole64(le64toh(f
->header
->n_data
) + 1);
1291 int journal_file_find_field_object_with_hash(
1293 const void *field
, uint64_t size
, uint64_t hash
,
1294 Object
**ret
, uint64_t *offset
) {
1296 uint64_t p
, osize
, h
, m
;
1301 assert(field
&& size
> 0);
1303 /* If the field hash table is empty, we can't find anything */
1304 if (le64toh(f
->header
->field_hash_table_size
) <= 0)
1307 /* Map the field hash table, if it isn't mapped yet. */
1308 r
= journal_file_map_field_hash_table(f
);
1312 osize
= offsetof(Object
, field
.payload
) + size
;
1314 m
= le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
);
1319 p
= le64toh(f
->field_hash_table
[h
].head_hash_offset
);
1324 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1328 if (le64toh(o
->field
.hash
) == hash
&&
1329 le64toh(o
->object
.size
) == osize
&&
1330 memcmp(o
->field
.payload
, field
, size
) == 0) {
1340 p
= le64toh(o
->field
.next_hash_offset
);
1346 int journal_file_find_field_object(
1348 const void *field
, uint64_t size
,
1349 Object
**ret
, uint64_t *offset
) {
1354 assert(field
&& size
> 0);
1356 hash
= hash64(field
, size
);
1358 return journal_file_find_field_object_with_hash(f
,
1363 int journal_file_find_data_object_with_hash(
1365 const void *data
, uint64_t size
, uint64_t hash
,
1366 Object
**ret
, uint64_t *offset
) {
1368 uint64_t p
, osize
, h
, m
;
1373 assert(data
|| size
== 0);
1375 /* If there's no data hash table, then there's no entry. */
1376 if (le64toh(f
->header
->data_hash_table_size
) <= 0)
1379 /* Map the data hash table, if it isn't mapped yet. */
1380 r
= journal_file_map_data_hash_table(f
);
1384 osize
= offsetof(Object
, data
.payload
) + size
;
1386 m
= le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
);
1391 p
= le64toh(f
->data_hash_table
[h
].head_hash_offset
);
1396 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1400 if (le64toh(o
->data
.hash
) != hash
)
1403 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
) {
1404 #if HAVE_XZ || HAVE_LZ4
1408 l
= le64toh(o
->object
.size
);
1409 if (l
<= offsetof(Object
, data
.payload
))
1412 l
-= offsetof(Object
, data
.payload
);
1414 r
= decompress_blob(o
->object
.flags
& OBJECT_COMPRESSION_MASK
,
1415 o
->data
.payload
, l
, &f
->compress_buffer
, &f
->compress_buffer_size
, &rsize
, 0);
1419 if (rsize
== size
&&
1420 memcmp(f
->compress_buffer
, data
, size
) == 0) {
1431 return -EPROTONOSUPPORT
;
1433 } else if (le64toh(o
->object
.size
) == osize
&&
1434 memcmp(o
->data
.payload
, data
, size
) == 0) {
1446 p
= le64toh(o
->data
.next_hash_offset
);
1452 int journal_file_find_data_object(
1454 const void *data
, uint64_t size
,
1455 Object
**ret
, uint64_t *offset
) {
1460 assert(data
|| size
== 0);
1462 hash
= hash64(data
, size
);
1464 return journal_file_find_data_object_with_hash(f
,
1469 static int journal_file_append_field(
1471 const void *field
, uint64_t size
,
1472 Object
**ret
, uint64_t *offset
) {
1480 assert(field
&& size
> 0);
1482 hash
= hash64(field
, size
);
1484 r
= journal_file_find_field_object_with_hash(f
, field
, size
, hash
, &o
, &p
);
1498 osize
= offsetof(Object
, field
.payload
) + size
;
1499 r
= journal_file_append_object(f
, OBJECT_FIELD
, osize
, &o
, &p
);
1503 o
->field
.hash
= htole64(hash
);
1504 memcpy(o
->field
.payload
, field
, size
);
1506 r
= journal_file_link_field(f
, o
, p
, hash
);
1510 /* The linking might have altered the window, so let's
1511 * refresh our pointer */
1512 r
= journal_file_move_to_object(f
, OBJECT_FIELD
, p
, &o
);
1517 r
= journal_file_hmac_put_object(f
, OBJECT_FIELD
, o
, p
);
1531 static int journal_file_append_data(
1533 const void *data
, uint64_t size
,
1534 Object
**ret
, uint64_t *offset
) {
1539 int r
, compression
= 0;
1543 assert(data
|| size
== 0);
1545 hash
= hash64(data
, size
);
1547 r
= journal_file_find_data_object_with_hash(f
, data
, size
, hash
, &o
, &p
);
1561 osize
= offsetof(Object
, data
.payload
) + size
;
1562 r
= journal_file_append_object(f
, OBJECT_DATA
, osize
, &o
, &p
);
1566 o
->data
.hash
= htole64(hash
);
1568 #if HAVE_XZ || HAVE_LZ4
1569 if (JOURNAL_FILE_COMPRESS(f
) && size
>= COMPRESSION_SIZE_THRESHOLD
) {
1572 compression
= compress_blob(data
, size
, o
->data
.payload
, size
- 1, &rsize
);
1574 if (compression
>= 0) {
1575 o
->object
.size
= htole64(offsetof(Object
, data
.payload
) + rsize
);
1576 o
->object
.flags
|= compression
;
1578 log_debug("Compressed data object %"PRIu64
" -> %zu using %s",
1579 size
, rsize
, object_compressed_to_string(compression
));
1581 /* Compression didn't work, we don't really care why, let's continue without compression */
1586 if (compression
== 0)
1587 memcpy_safe(o
->data
.payload
, data
, size
);
1589 r
= journal_file_link_data(f
, o
, p
, hash
);
1594 r
= journal_file_hmac_put_object(f
, OBJECT_DATA
, o
, p
);
1599 /* The linking might have altered the window, so let's
1600 * refresh our pointer */
1601 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1608 eq
= memchr(data
, '=', size
);
1609 if (eq
&& eq
> data
) {
1613 /* Create field object ... */
1614 r
= journal_file_append_field(f
, data
, (uint8_t*) eq
- (uint8_t*) data
, &fo
, &fp
);
1618 /* ... and link it in. */
1619 o
->data
.next_field_offset
= fo
->field
.head_data_offset
;
1620 fo
->field
.head_data_offset
= le64toh(p
);
1632 uint64_t journal_file_entry_n_items(Object
*o
) {
1635 if (o
->object
.type
!= OBJECT_ENTRY
)
1638 return (le64toh(o
->object
.size
) - offsetof(Object
, entry
.items
)) / sizeof(EntryItem
);
1641 uint64_t journal_file_entry_array_n_items(Object
*o
) {
1644 if (o
->object
.type
!= OBJECT_ENTRY_ARRAY
)
1647 return (le64toh(o
->object
.size
) - offsetof(Object
, entry_array
.items
)) / sizeof(uint64_t);
1650 uint64_t journal_file_hash_table_n_items(Object
*o
) {
1653 if (!IN_SET(o
->object
.type
, OBJECT_DATA_HASH_TABLE
, OBJECT_FIELD_HASH_TABLE
))
1656 return (le64toh(o
->object
.size
) - offsetof(Object
, hash_table
.items
)) / sizeof(HashItem
);
1659 static int link_entry_into_array(JournalFile
*f
,
1664 uint64_t n
= 0, ap
= 0, q
, i
, a
, hidx
;
1673 a
= le64toh(*first
);
1674 i
= hidx
= le64toh(*idx
);
1677 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
1681 n
= journal_file_entry_array_n_items(o
);
1683 o
->entry_array
.items
[i
] = htole64(p
);
1684 *idx
= htole64(hidx
+ 1);
1690 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
1701 r
= journal_file_append_object(f
, OBJECT_ENTRY_ARRAY
,
1702 offsetof(Object
, entry_array
.items
) + n
* sizeof(uint64_t),
1708 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY_ARRAY
, o
, q
);
1713 o
->entry_array
.items
[i
] = htole64(p
);
1716 *first
= htole64(q
);
1718 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, ap
, &o
);
1722 o
->entry_array
.next_entry_array_offset
= htole64(q
);
1725 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
1726 f
->header
->n_entry_arrays
= htole64(le64toh(f
->header
->n_entry_arrays
) + 1);
1728 *idx
= htole64(hidx
+ 1);
1733 static int link_entry_into_array_plus_one(JournalFile
*f
,
1748 *extra
= htole64(p
);
1752 i
= htole64(le64toh(*idx
) - 1);
1753 r
= link_entry_into_array(f
, first
, &i
, p
);
1758 *idx
= htole64(le64toh(*idx
) + 1);
1762 static int journal_file_link_entry_item(JournalFile
*f
, Object
*o
, uint64_t offset
, uint64_t i
) {
1769 p
= le64toh(o
->entry
.items
[i
].object_offset
);
1773 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
1777 return link_entry_into_array_plus_one(f
,
1778 &o
->data
.entry_offset
,
1779 &o
->data
.entry_array_offset
,
1784 static int journal_file_link_entry(JournalFile
*f
, Object
*o
, uint64_t offset
) {
1793 if (o
->object
.type
!= OBJECT_ENTRY
)
1796 __sync_synchronize();
1798 /* Link up the entry itself */
1799 r
= link_entry_into_array(f
,
1800 &f
->header
->entry_array_offset
,
1801 &f
->header
->n_entries
,
1806 /* log_debug("=> %s seqnr=%"PRIu64" n_entries=%"PRIu64, f->path, o->entry.seqnum, f->header->n_entries); */
1808 if (f
->header
->head_entry_realtime
== 0)
1809 f
->header
->head_entry_realtime
= o
->entry
.realtime
;
1811 f
->header
->tail_entry_realtime
= o
->entry
.realtime
;
1812 f
->header
->tail_entry_monotonic
= o
->entry
.monotonic
;
1814 f
->tail_entry_monotonic_valid
= true;
1816 /* Link up the items */
1817 n
= journal_file_entry_n_items(o
);
1818 for (i
= 0; i
< n
; i
++) {
1819 r
= journal_file_link_entry_item(f
, o
, offset
, i
);
1827 static int journal_file_append_entry_internal(
1829 const dual_timestamp
*ts
,
1831 const EntryItem items
[], unsigned n_items
,
1833 Object
**ret
, uint64_t *offset
) {
1841 assert(items
|| n_items
== 0);
1844 osize
= offsetof(Object
, entry
.items
) + (n_items
* sizeof(EntryItem
));
1846 r
= journal_file_append_object(f
, OBJECT_ENTRY
, osize
, &o
, &np
);
1850 o
->entry
.seqnum
= htole64(journal_file_entry_seqnum(f
, seqnum
));
1851 memcpy_safe(o
->entry
.items
, items
, n_items
* sizeof(EntryItem
));
1852 o
->entry
.realtime
= htole64(ts
->realtime
);
1853 o
->entry
.monotonic
= htole64(ts
->monotonic
);
1854 o
->entry
.xor_hash
= htole64(xor_hash
);
1855 o
->entry
.boot_id
= f
->header
->boot_id
;
1858 r
= journal_file_hmac_put_object(f
, OBJECT_ENTRY
, o
, np
);
1863 r
= journal_file_link_entry(f
, o
, np
);
1876 void journal_file_post_change(JournalFile
*f
) {
1879 /* inotify() does not receive IN_MODIFY events from file
1880 * accesses done via mmap(). After each access we hence
1881 * trigger IN_MODIFY by truncating the journal file to its
1882 * current size which triggers IN_MODIFY. */
1884 __sync_synchronize();
1886 if (ftruncate(f
->fd
, f
->last_stat
.st_size
) < 0)
1887 log_debug_errno(errno
, "Failed to truncate file to its own size: %m");
1890 static int post_change_thunk(sd_event_source
*timer
, uint64_t usec
, void *userdata
) {
1893 journal_file_post_change(userdata
);
1898 static void schedule_post_change(JournalFile
*f
) {
1899 sd_event_source
*timer
;
1904 assert(f
->post_change_timer
);
1906 timer
= f
->post_change_timer
;
1908 r
= sd_event_source_get_enabled(timer
, &enabled
);
1910 log_debug_errno(r
, "Failed to get ftruncate timer state: %m");
1914 if (enabled
== SD_EVENT_ONESHOT
)
1917 r
= sd_event_now(sd_event_source_get_event(timer
), CLOCK_MONOTONIC
, &now
);
1919 log_debug_errno(r
, "Failed to get clock's now for scheduling ftruncate: %m");
1923 r
= sd_event_source_set_time(timer
, now
+f
->post_change_timer_period
);
1925 log_debug_errno(r
, "Failed to set time for scheduling ftruncate: %m");
1929 r
= sd_event_source_set_enabled(timer
, SD_EVENT_ONESHOT
);
1931 log_debug_errno(r
, "Failed to enable scheduled ftruncate: %m");
1938 /* On failure, let's simply post the change immediately. */
1939 journal_file_post_change(f
);
1942 /* Enable coalesced change posting in a timer on the provided sd_event instance */
1943 int journal_file_enable_post_change_timer(JournalFile
*f
, sd_event
*e
, usec_t t
) {
1944 _cleanup_(sd_event_source_unrefp
) sd_event_source
*timer
= NULL
;
1948 assert_return(!f
->post_change_timer
, -EINVAL
);
1952 r
= sd_event_add_time(e
, &timer
, CLOCK_MONOTONIC
, 0, 0, post_change_thunk
, f
);
1956 r
= sd_event_source_set_enabled(timer
, SD_EVENT_OFF
);
1960 f
->post_change_timer
= timer
;
1962 f
->post_change_timer_period
= t
;
1967 static int entry_item_cmp(const void *_a
, const void *_b
) {
1968 const EntryItem
*a
= _a
, *b
= _b
;
1970 if (le64toh(a
->object_offset
) < le64toh(b
->object_offset
))
1972 if (le64toh(a
->object_offset
) > le64toh(b
->object_offset
))
1977 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
) {
1981 uint64_t xor_hash
= 0;
1982 struct dual_timestamp _ts
;
1986 assert(iovec
|| n_iovec
== 0);
1989 dual_timestamp_get(&_ts
);
1994 r
= journal_file_maybe_append_tag(f
, ts
->realtime
);
1999 /* alloca() can't take 0, hence let's allocate at least one */
2000 items
= alloca(sizeof(EntryItem
) * MAX(1u, n_iovec
));
2002 for (i
= 0; i
< n_iovec
; i
++) {
2006 r
= journal_file_append_data(f
, iovec
[i
].iov_base
, iovec
[i
].iov_len
, &o
, &p
);
2010 xor_hash
^= le64toh(o
->data
.hash
);
2011 items
[i
].object_offset
= htole64(p
);
2012 items
[i
].hash
= o
->data
.hash
;
2015 /* Order by the position on disk, in order to improve seek
2016 * times for rotating media. */
2017 qsort_safe(items
, n_iovec
, sizeof(EntryItem
), entry_item_cmp
);
2019 r
= journal_file_append_entry_internal(f
, ts
, xor_hash
, items
, n_iovec
, seqnum
, ret
, offset
);
2021 /* If the memory mapping triggered a SIGBUS then we return an
2022 * IO error and ignore the error code passed down to us, since
2023 * it is very likely just an effect of a nullified replacement
2026 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
2029 if (f
->post_change_timer
)
2030 schedule_post_change(f
);
2032 journal_file_post_change(f
);
2037 typedef struct ChainCacheItem
{
2038 uint64_t first
; /* the array at the beginning of the chain */
2039 uint64_t array
; /* the cached array */
2040 uint64_t begin
; /* the first item in the cached array */
2041 uint64_t total
; /* the total number of items in all arrays before this one in the chain */
2042 uint64_t last_index
; /* the last index we looked at, to optimize locality when bisecting */
2045 static void chain_cache_put(
2052 uint64_t last_index
) {
2055 /* If the chain item to cache for this chain is the
2056 * first one it's not worth caching anything */
2060 if (ordered_hashmap_size(h
) >= CHAIN_CACHE_MAX
) {
2061 ci
= ordered_hashmap_steal_first(h
);
2064 ci
= new(ChainCacheItem
, 1);
2071 if (ordered_hashmap_put(h
, &ci
->first
, ci
) < 0) {
2076 assert(ci
->first
== first
);
2081 ci
->last_index
= last_index
;
2084 static int generic_array_get(
2088 Object
**ret
, uint64_t *offset
) {
2091 uint64_t p
= 0, a
, t
= 0;
2099 /* Try the chain cache first */
2100 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
2101 if (ci
&& i
> ci
->total
) {
2110 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &o
);
2114 k
= journal_file_entry_array_n_items(o
);
2116 p
= le64toh(o
->entry_array
.items
[i
]);
2122 a
= le64toh(o
->entry_array
.next_entry_array_offset
);
2128 /* Let's cache this item for the next invocation */
2129 chain_cache_put(f
->chain_cache
, ci
, first
, a
, le64toh(o
->entry_array
.items
[0]), t
, i
);
2131 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2144 static int generic_array_get_plus_one(
2149 Object
**ret
, uint64_t *offset
) {
2158 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, &o
);
2171 return generic_array_get(f
, first
, i
-1, ret
, offset
);
2180 static int generic_array_bisect(
2185 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
2186 direction_t direction
,
2191 uint64_t a
, p
, t
= 0, i
= 0, last_p
= 0, last_index
= (uint64_t) -1;
2192 bool subtract_one
= false;
2193 Object
*o
, *array
= NULL
;
2198 assert(test_object
);
2200 /* Start with the first array in the chain */
2203 ci
= ordered_hashmap_get(f
->chain_cache
, &first
);
2204 if (ci
&& n
> ci
->total
) {
2205 /* Ah, we have iterated this bisection array chain
2206 * previously! Let's see if we can skip ahead in the
2207 * chain, as far as the last time. But we can't jump
2208 * backwards in the chain, so let's check that
2211 r
= test_object(f
, ci
->begin
, needle
);
2215 if (r
== TEST_LEFT
) {
2216 /* OK, what we are looking for is right of the
2217 * begin of this EntryArray, so let's jump
2218 * straight to previously cached array in the
2224 last_index
= ci
->last_index
;
2229 uint64_t left
, right
, k
, lp
;
2231 r
= journal_file_move_to_object(f
, OBJECT_ENTRY_ARRAY
, a
, &array
);
2235 k
= journal_file_entry_array_n_items(array
);
2241 lp
= p
= le64toh(array
->entry_array
.items
[i
]);
2245 r
= test_object(f
, p
, needle
);
2246 if (r
== -EBADMSG
) {
2247 log_debug_errno(r
, "Encountered invalid entry while bisecting, cutting algorithm short. (1)");
2254 if (r
== TEST_FOUND
)
2255 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2257 if (r
== TEST_RIGHT
) {
2261 if (last_index
!= (uint64_t) -1) {
2262 assert(last_index
<= right
);
2264 /* If we cached the last index we
2265 * looked at, let's try to not to jump
2266 * too wildly around and see if we can
2267 * limit the range to look at early to
2268 * the immediate neighbors of the last
2269 * index we looked at. */
2271 if (last_index
> 0) {
2272 uint64_t x
= last_index
- 1;
2274 p
= le64toh(array
->entry_array
.items
[x
]);
2278 r
= test_object(f
, p
, needle
);
2282 if (r
== TEST_FOUND
)
2283 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2285 if (r
== TEST_RIGHT
)
2291 if (last_index
< right
) {
2292 uint64_t y
= last_index
+ 1;
2294 p
= le64toh(array
->entry_array
.items
[y
]);
2298 r
= test_object(f
, p
, needle
);
2302 if (r
== TEST_FOUND
)
2303 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2305 if (r
== TEST_RIGHT
)
2313 if (left
== right
) {
2314 if (direction
== DIRECTION_UP
)
2315 subtract_one
= true;
2321 assert(left
< right
);
2322 i
= (left
+ right
) / 2;
2324 p
= le64toh(array
->entry_array
.items
[i
]);
2328 r
= test_object(f
, p
, needle
);
2329 if (r
== -EBADMSG
) {
2330 log_debug_errno(r
, "Encountered invalid entry while bisecting, cutting algorithm short. (2)");
2337 if (r
== TEST_FOUND
)
2338 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2340 if (r
== TEST_RIGHT
)
2348 if (direction
== DIRECTION_UP
) {
2350 subtract_one
= true;
2361 last_index
= (uint64_t) -1;
2362 a
= le64toh(array
->entry_array
.next_entry_array_offset
);
2368 if (subtract_one
&& t
== 0 && i
== 0)
2371 /* Let's cache this item for the next invocation */
2372 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
);
2374 if (subtract_one
&& i
== 0)
2376 else if (subtract_one
)
2377 p
= le64toh(array
->entry_array
.items
[i
-1]);
2379 p
= le64toh(array
->entry_array
.items
[i
]);
2381 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2392 *idx
= t
+ i
+ (subtract_one
? -1 : 0);
2397 static int generic_array_bisect_plus_one(
2403 int (*test_object
)(JournalFile
*f
, uint64_t p
, uint64_t needle
),
2404 direction_t direction
,
2410 bool step_back
= false;
2414 assert(test_object
);
2419 /* This bisects the array in object 'first', but first checks
2421 r
= test_object(f
, extra
, needle
);
2425 if (r
== TEST_FOUND
)
2426 r
= direction
== DIRECTION_DOWN
? TEST_RIGHT
: TEST_LEFT
;
2428 /* if we are looking with DIRECTION_UP then we need to first
2429 see if in the actual array there is a matching entry, and
2430 return the last one of that. But if there isn't any we need
2431 to return this one. Hence remember this, and return it
2434 step_back
= direction
== DIRECTION_UP
;
2436 if (r
== TEST_RIGHT
) {
2437 if (direction
== DIRECTION_DOWN
)
2443 r
= generic_array_bisect(f
, first
, n
-1, needle
, test_object
, direction
, ret
, offset
, idx
);
2445 if (r
== 0 && step_back
)
2454 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, extra
, &o
);
2470 _pure_
static int test_object_offset(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2476 else if (p
< needle
)
2482 static int test_object_seqnum(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2489 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2493 if (le64toh(o
->entry
.seqnum
) == needle
)
2495 else if (le64toh(o
->entry
.seqnum
) < needle
)
2501 int journal_file_move_to_entry_by_seqnum(
2504 direction_t direction
,
2510 return generic_array_bisect(f
,
2511 le64toh(f
->header
->entry_array_offset
),
2512 le64toh(f
->header
->n_entries
),
2519 static int test_object_realtime(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2526 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2530 if (le64toh(o
->entry
.realtime
) == needle
)
2532 else if (le64toh(o
->entry
.realtime
) < needle
)
2538 int journal_file_move_to_entry_by_realtime(
2541 direction_t direction
,
2547 return generic_array_bisect(f
,
2548 le64toh(f
->header
->entry_array_offset
),
2549 le64toh(f
->header
->n_entries
),
2551 test_object_realtime
,
2556 static int test_object_monotonic(JournalFile
*f
, uint64_t p
, uint64_t needle
) {
2563 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, p
, &o
);
2567 if (le64toh(o
->entry
.monotonic
) == needle
)
2569 else if (le64toh(o
->entry
.monotonic
) < needle
)
2575 static int find_data_object_by_boot_id(
2581 char t
[sizeof("_BOOT_ID=")-1 + 32 + 1] = "_BOOT_ID=";
2583 sd_id128_to_string(boot_id
, t
+ 9);
2584 return journal_file_find_data_object(f
, t
, sizeof(t
) - 1, o
, b
);
2587 int journal_file_move_to_entry_by_monotonic(
2591 direction_t direction
,
2600 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, NULL
);
2606 return generic_array_bisect_plus_one(f
,
2607 le64toh(o
->data
.entry_offset
),
2608 le64toh(o
->data
.entry_array_offset
),
2609 le64toh(o
->data
.n_entries
),
2611 test_object_monotonic
,
2616 void journal_file_reset_location(JournalFile
*f
) {
2617 f
->location_type
= LOCATION_HEAD
;
2618 f
->current_offset
= 0;
2619 f
->current_seqnum
= 0;
2620 f
->current_realtime
= 0;
2621 f
->current_monotonic
= 0;
2622 zero(f
->current_boot_id
);
2623 f
->current_xor_hash
= 0;
2626 void journal_file_save_location(JournalFile
*f
, Object
*o
, uint64_t offset
) {
2627 f
->location_type
= LOCATION_SEEK
;
2628 f
->current_offset
= offset
;
2629 f
->current_seqnum
= le64toh(o
->entry
.seqnum
);
2630 f
->current_realtime
= le64toh(o
->entry
.realtime
);
2631 f
->current_monotonic
= le64toh(o
->entry
.monotonic
);
2632 f
->current_boot_id
= o
->entry
.boot_id
;
2633 f
->current_xor_hash
= le64toh(o
->entry
.xor_hash
);
2636 int journal_file_compare_locations(JournalFile
*af
, JournalFile
*bf
) {
2641 assert(af
->location_type
== LOCATION_SEEK
);
2642 assert(bf
->location_type
== LOCATION_SEEK
);
2644 /* If contents and timestamps match, these entries are
2645 * identical, even if the seqnum does not match */
2646 if (sd_id128_equal(af
->current_boot_id
, bf
->current_boot_id
) &&
2647 af
->current_monotonic
== bf
->current_monotonic
&&
2648 af
->current_realtime
== bf
->current_realtime
&&
2649 af
->current_xor_hash
== bf
->current_xor_hash
)
2652 if (sd_id128_equal(af
->header
->seqnum_id
, bf
->header
->seqnum_id
)) {
2654 /* If this is from the same seqnum source, compare
2656 if (af
->current_seqnum
< bf
->current_seqnum
)
2658 if (af
->current_seqnum
> bf
->current_seqnum
)
2661 /* Wow! This is weird, different data but the same
2662 * seqnums? Something is borked, but let's make the
2663 * best of it and compare by time. */
2666 if (sd_id128_equal(af
->current_boot_id
, bf
->current_boot_id
)) {
2668 /* If the boot id matches, compare monotonic time */
2669 if (af
->current_monotonic
< bf
->current_monotonic
)
2671 if (af
->current_monotonic
> bf
->current_monotonic
)
2675 /* Otherwise, compare UTC time */
2676 if (af
->current_realtime
< bf
->current_realtime
)
2678 if (af
->current_realtime
> bf
->current_realtime
)
2681 /* Finally, compare by contents */
2682 if (af
->current_xor_hash
< bf
->current_xor_hash
)
2684 if (af
->current_xor_hash
> bf
->current_xor_hash
)
2690 static int bump_array_index(uint64_t *i
, direction_t direction
, uint64_t n
) {
2692 /* Increase or decrease the specified index, in the right direction. */
2694 if (direction
== DIRECTION_DOWN
) {
2709 static bool check_properly_ordered(uint64_t new_offset
, uint64_t old_offset
, direction_t direction
) {
2711 /* Consider it an error if any of the two offsets is uninitialized */
2712 if (old_offset
== 0 || new_offset
== 0)
2715 /* If we go down, the new offset must be larger than the old one. */
2716 return direction
== DIRECTION_DOWN
?
2717 new_offset
> old_offset
:
2718 new_offset
< old_offset
;
2721 int journal_file_next_entry(
2724 direction_t direction
,
2725 Object
**ret
, uint64_t *offset
) {
2733 n
= le64toh(f
->header
->n_entries
);
2738 i
= direction
== DIRECTION_DOWN
? 0 : n
- 1;
2740 r
= generic_array_bisect(f
,
2741 le64toh(f
->header
->entry_array_offset
),
2742 le64toh(f
->header
->n_entries
),
2751 r
= bump_array_index(&i
, direction
, n
);
2756 /* And jump to it */
2758 r
= generic_array_get(f
,
2759 le64toh(f
->header
->entry_array_offset
),
2767 /* OK, so this entry is borked. Most likely some entry didn't get synced to disk properly, let's see if
2768 * the next one might work for us instead. */
2769 log_debug_errno(r
, "Entry item %" PRIu64
" is bad, skipping over it.", i
);
2771 r
= bump_array_index(&i
, direction
, n
);
2776 /* Ensure our array is properly ordered. */
2777 if (p
> 0 && !check_properly_ordered(ofs
, p
, direction
)) {
2778 log_debug("%s: entry array not properly ordered at entry %" PRIu64
, f
->path
, i
);
2788 int journal_file_next_entry_for_data(
2790 Object
*o
, uint64_t p
,
2791 uint64_t data_offset
,
2792 direction_t direction
,
2793 Object
**ret
, uint64_t *offset
) {
2800 assert(p
> 0 || !o
);
2802 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2806 n
= le64toh(d
->data
.n_entries
);
2811 i
= direction
== DIRECTION_DOWN
? 0 : n
- 1;
2813 if (o
->object
.type
!= OBJECT_ENTRY
)
2816 r
= generic_array_bisect_plus_one(f
,
2817 le64toh(d
->data
.entry_offset
),
2818 le64toh(d
->data
.entry_array_offset
),
2819 le64toh(d
->data
.n_entries
),
2829 r
= bump_array_index(&i
, direction
, n
);
2835 r
= generic_array_get_plus_one(f
,
2836 le64toh(d
->data
.entry_offset
),
2837 le64toh(d
->data
.entry_array_offset
),
2845 log_debug_errno(r
, "Data entry item %" PRIu64
" is bad, skipping over it.", i
);
2847 r
= bump_array_index(&i
, direction
, n
);
2852 /* Ensure our array is properly ordered. */
2853 if (p
> 0 && check_properly_ordered(ofs
, p
, direction
)) {
2854 log_debug("%s data entry array not properly ordered at entry %" PRIu64
, f
->path
, i
);
2864 int journal_file_move_to_entry_by_offset_for_data(
2866 uint64_t data_offset
,
2868 direction_t direction
,
2869 Object
**ret
, uint64_t *offset
) {
2876 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2880 return generic_array_bisect_plus_one(f
,
2881 le64toh(d
->data
.entry_offset
),
2882 le64toh(d
->data
.entry_array_offset
),
2883 le64toh(d
->data
.n_entries
),
2890 int journal_file_move_to_entry_by_monotonic_for_data(
2892 uint64_t data_offset
,
2895 direction_t direction
,
2896 Object
**ret
, uint64_t *offset
) {
2904 /* First, seek by time */
2905 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, &b
);
2911 r
= generic_array_bisect_plus_one(f
,
2912 le64toh(o
->data
.entry_offset
),
2913 le64toh(o
->data
.entry_array_offset
),
2914 le64toh(o
->data
.n_entries
),
2916 test_object_monotonic
,
2922 /* And now, continue seeking until we find an entry that
2923 * exists in both bisection arrays */
2929 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2933 r
= generic_array_bisect_plus_one(f
,
2934 le64toh(d
->data
.entry_offset
),
2935 le64toh(d
->data
.entry_array_offset
),
2936 le64toh(d
->data
.n_entries
),
2944 r
= journal_file_move_to_object(f
, OBJECT_DATA
, b
, &o
);
2948 r
= generic_array_bisect_plus_one(f
,
2949 le64toh(o
->data
.entry_offset
),
2950 le64toh(o
->data
.entry_array_offset
),
2951 le64toh(o
->data
.n_entries
),
2973 int journal_file_move_to_entry_by_seqnum_for_data(
2975 uint64_t data_offset
,
2977 direction_t direction
,
2978 Object
**ret
, uint64_t *offset
) {
2985 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
2989 return generic_array_bisect_plus_one(f
,
2990 le64toh(d
->data
.entry_offset
),
2991 le64toh(d
->data
.entry_array_offset
),
2992 le64toh(d
->data
.n_entries
),
2999 int journal_file_move_to_entry_by_realtime_for_data(
3001 uint64_t data_offset
,
3003 direction_t direction
,
3004 Object
**ret
, uint64_t *offset
) {
3011 r
= journal_file_move_to_object(f
, OBJECT_DATA
, data_offset
, &d
);
3015 return generic_array_bisect_plus_one(f
,
3016 le64toh(d
->data
.entry_offset
),
3017 le64toh(d
->data
.entry_array_offset
),
3018 le64toh(d
->data
.n_entries
),
3020 test_object_realtime
,
3025 void journal_file_dump(JournalFile
*f
) {
3033 journal_file_print_header(f
);
3035 p
= le64toh(f
->header
->header_size
);
3037 r
= journal_file_move_to_object(f
, OBJECT_UNUSED
, p
, &o
);
3041 switch (o
->object
.type
) {
3044 printf("Type: OBJECT_UNUSED\n");
3048 printf("Type: OBJECT_DATA\n");
3052 printf("Type: OBJECT_FIELD\n");
3056 printf("Type: OBJECT_ENTRY seqnum=%"PRIu64
" monotonic=%"PRIu64
" realtime=%"PRIu64
"\n",
3057 le64toh(o
->entry
.seqnum
),
3058 le64toh(o
->entry
.monotonic
),
3059 le64toh(o
->entry
.realtime
));
3062 case OBJECT_FIELD_HASH_TABLE
:
3063 printf("Type: OBJECT_FIELD_HASH_TABLE\n");
3066 case OBJECT_DATA_HASH_TABLE
:
3067 printf("Type: OBJECT_DATA_HASH_TABLE\n");
3070 case OBJECT_ENTRY_ARRAY
:
3071 printf("Type: OBJECT_ENTRY_ARRAY\n");
3075 printf("Type: OBJECT_TAG seqnum=%"PRIu64
" epoch=%"PRIu64
"\n",
3076 le64toh(o
->tag
.seqnum
),
3077 le64toh(o
->tag
.epoch
));
3081 printf("Type: unknown (%i)\n", o
->object
.type
);
3085 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
)
3086 printf("Flags: %s\n",
3087 object_compressed_to_string(o
->object
.flags
& OBJECT_COMPRESSION_MASK
));
3089 if (p
== le64toh(f
->header
->tail_object_offset
))
3092 p
= p
+ ALIGN64(le64toh(o
->object
.size
));
3097 log_error("File corrupt");
3100 static const char* format_timestamp_safe(char *buf
, size_t l
, usec_t t
) {
3103 x
= format_timestamp(buf
, l
, t
);
3109 void journal_file_print_header(JournalFile
*f
) {
3110 char a
[33], b
[33], c
[33], d
[33];
3111 char x
[FORMAT_TIMESTAMP_MAX
], y
[FORMAT_TIMESTAMP_MAX
], z
[FORMAT_TIMESTAMP_MAX
];
3113 char bytes
[FORMAT_BYTES_MAX
];
3118 printf("File Path: %s\n"
3122 "Sequential Number ID: %s\n"
3124 "Compatible Flags:%s%s\n"
3125 "Incompatible Flags:%s%s%s\n"
3126 "Header size: %"PRIu64
"\n"
3127 "Arena size: %"PRIu64
"\n"
3128 "Data Hash Table Size: %"PRIu64
"\n"
3129 "Field Hash Table Size: %"PRIu64
"\n"
3130 "Rotate Suggested: %s\n"
3131 "Head Sequential Number: %"PRIu64
" (%"PRIx64
")\n"
3132 "Tail Sequential Number: %"PRIu64
" (%"PRIx64
")\n"
3133 "Head Realtime Timestamp: %s (%"PRIx64
")\n"
3134 "Tail Realtime Timestamp: %s (%"PRIx64
")\n"
3135 "Tail Monotonic Timestamp: %s (%"PRIx64
")\n"
3136 "Objects: %"PRIu64
"\n"
3137 "Entry Objects: %"PRIu64
"\n",
3139 sd_id128_to_string(f
->header
->file_id
, a
),
3140 sd_id128_to_string(f
->header
->machine_id
, b
),
3141 sd_id128_to_string(f
->header
->boot_id
, c
),
3142 sd_id128_to_string(f
->header
->seqnum_id
, d
),
3143 f
->header
->state
== STATE_OFFLINE
? "OFFLINE" :
3144 f
->header
->state
== STATE_ONLINE
? "ONLINE" :
3145 f
->header
->state
== STATE_ARCHIVED
? "ARCHIVED" : "UNKNOWN",
3146 JOURNAL_HEADER_SEALED(f
->header
) ? " SEALED" : "",
3147 (le32toh(f
->header
->compatible_flags
) & ~HEADER_COMPATIBLE_ANY
) ? " ???" : "",
3148 JOURNAL_HEADER_COMPRESSED_XZ(f
->header
) ? " COMPRESSED-XZ" : "",
3149 JOURNAL_HEADER_COMPRESSED_LZ4(f
->header
) ? " COMPRESSED-LZ4" : "",
3150 (le32toh(f
->header
->incompatible_flags
) & ~HEADER_INCOMPATIBLE_ANY
) ? " ???" : "",
3151 le64toh(f
->header
->header_size
),
3152 le64toh(f
->header
->arena_size
),
3153 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
3154 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
),
3155 yes_no(journal_file_rotate_suggested(f
, 0)),
3156 le64toh(f
->header
->head_entry_seqnum
), le64toh(f
->header
->head_entry_seqnum
),
3157 le64toh(f
->header
->tail_entry_seqnum
), le64toh(f
->header
->tail_entry_seqnum
),
3158 format_timestamp_safe(x
, sizeof(x
), le64toh(f
->header
->head_entry_realtime
)), le64toh(f
->header
->head_entry_realtime
),
3159 format_timestamp_safe(y
, sizeof(y
), le64toh(f
->header
->tail_entry_realtime
)), le64toh(f
->header
->tail_entry_realtime
),
3160 format_timespan(z
, sizeof(z
), le64toh(f
->header
->tail_entry_monotonic
), USEC_PER_MSEC
), le64toh(f
->header
->tail_entry_monotonic
),
3161 le64toh(f
->header
->n_objects
),
3162 le64toh(f
->header
->n_entries
));
3164 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
3165 printf("Data Objects: %"PRIu64
"\n"
3166 "Data Hash Table Fill: %.1f%%\n",
3167 le64toh(f
->header
->n_data
),
3168 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))));
3170 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
3171 printf("Field Objects: %"PRIu64
"\n"
3172 "Field Hash Table Fill: %.1f%%\n",
3173 le64toh(f
->header
->n_fields
),
3174 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))));
3176 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_tags
))
3177 printf("Tag Objects: %"PRIu64
"\n",
3178 le64toh(f
->header
->n_tags
));
3179 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_entry_arrays
))
3180 printf("Entry Array Objects: %"PRIu64
"\n",
3181 le64toh(f
->header
->n_entry_arrays
));
3183 if (fstat(f
->fd
, &st
) >= 0)
3184 printf("Disk usage: %s\n", format_bytes(bytes
, sizeof(bytes
), (uint64_t) st
.st_blocks
* 512ULL));
3187 static int journal_file_warn_btrfs(JournalFile
*f
) {
3193 /* Before we write anything, check if the COW logic is turned
3194 * off on btrfs. Given our write pattern that is quite
3195 * unfriendly to COW file systems this should greatly improve
3196 * performance on COW file systems, such as btrfs, at the
3197 * expense of data integrity features (which shouldn't be too
3198 * bad, given that we do our own checksumming). */
3200 r
= btrfs_is_filesystem(f
->fd
);
3202 return log_warning_errno(r
, "Failed to determine if journal is on btrfs: %m");
3206 r
= read_attr_fd(f
->fd
, &attrs
);
3208 return log_warning_errno(r
, "Failed to read file attributes: %m");
3210 if (attrs
& FS_NOCOW_FL
) {
3211 log_debug("Detected btrfs file system with copy-on-write disabled, all is good.");
3215 log_notice("Creating journal file %s on a btrfs file system, and copy-on-write is enabled. "
3216 "This is likely to slow down journal access substantially, please consider turning "
3217 "off the copy-on-write file attribute on the journal directory, using chattr +C.", f
->path
);
3222 int journal_file_open(
3229 JournalMetrics
*metrics
,
3230 MMapCache
*mmap_cache
,
3231 Set
*deferred_closes
,
3232 JournalFile
*template,
3233 JournalFile
**ret
) {
3235 bool newly_created
= false;
3241 assert(fd
>= 0 || fname
);
3243 if (!IN_SET((flags
& O_ACCMODE
), O_RDONLY
, O_RDWR
))
3247 if (!endswith(fname
, ".journal") &&
3248 !endswith(fname
, ".journal~"))
3252 f
= new0(JournalFile
, 1);
3260 f
->prot
= prot_from_flags(flags
);
3261 f
->writable
= (flags
& O_ACCMODE
) != O_RDONLY
;
3263 f
->compress_lz4
= compress
;
3265 f
->compress_xz
= compress
;
3272 f
->mmap
= mmap_cache_ref(mmap_cache
);
3274 f
->mmap
= mmap_cache_new();
3282 f
->path
= strdup(fname
);
3288 /* If we don't know the path, fill in something explanatory and vaguely useful */
3289 if (asprintf(&f
->path
, "/proc/self/%i", fd
) < 0) {
3295 f
->chain_cache
= ordered_hashmap_new(&uint64_hash_ops
);
3296 if (!f
->chain_cache
) {
3302 f
->fd
= open(f
->path
, f
->flags
|O_CLOEXEC
, f
->mode
);
3308 /* fds we opened here by us should also be closed by us. */
3312 f
->cache_fd
= mmap_cache_add_fd(f
->mmap
, f
->fd
);
3318 r
= journal_file_fstat(f
);
3322 if (f
->last_stat
.st_size
== 0 && f
->writable
) {
3324 (void) journal_file_warn_btrfs(f
);
3326 /* Let's attach the creation time to the journal file,
3327 * so that the vacuuming code knows the age of this
3328 * file even if the file might end up corrupted one
3329 * day... Ideally we'd just use the creation time many
3330 * file systems maintain for each file, but there is
3331 * currently no usable API to query this, hence let's
3332 * emulate this via extended attributes. If extended
3333 * attributes are not supported we'll just skip this,
3334 * and rely solely on mtime/atime/ctime of the file. */
3336 fd_setcrtime(f
->fd
, 0);
3339 /* Try to load the FSPRG state, and if we can't, then
3340 * just don't do sealing */
3342 r
= journal_file_fss_load(f
);
3348 r
= journal_file_init_header(f
, template);
3352 r
= journal_file_fstat(f
);
3356 newly_created
= true;
3359 if (f
->last_stat
.st_size
< (off_t
) HEADER_SIZE_MIN
) {
3364 r
= mmap_cache_get(f
->mmap
, f
->cache_fd
, f
->prot
, CONTEXT_HEADER
, true, 0, PAGE_ALIGN(sizeof(Header
)), &f
->last_stat
, &h
, NULL
);
3370 if (!newly_created
) {
3371 if (deferred_closes
)
3372 journal_file_close_set(deferred_closes
);
3374 r
= journal_file_verify_header(f
);
3380 if (!newly_created
&& f
->writable
) {
3381 r
= journal_file_fss_load(f
);
3389 journal_default_metrics(metrics
, f
->fd
);
3390 f
->metrics
= *metrics
;
3391 } else if (template)
3392 f
->metrics
= template->metrics
;
3394 r
= journal_file_refresh_header(f
);
3400 r
= journal_file_hmac_setup(f
);
3405 if (newly_created
) {
3406 r
= journal_file_setup_field_hash_table(f
);
3410 r
= journal_file_setup_data_hash_table(f
);
3415 r
= journal_file_append_first_tag(f
);
3421 if (mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
)) {
3426 if (template && template->post_change_timer
) {
3427 r
= journal_file_enable_post_change_timer(
3429 sd_event_source_get_event(template->post_change_timer
),
3430 template->post_change_timer_period
);
3436 /* The file is opened now successfully, thus we take possession of any passed in fd. */
3443 if (f
->cache_fd
&& mmap_cache_got_sigbus(f
->mmap
, f
->cache_fd
))
3446 (void) journal_file_close(f
);
3451 int journal_file_rotate(JournalFile
**f
, bool compress
, bool seal
, Set
*deferred_closes
) {
3452 _cleanup_free_
char *p
= NULL
;
3454 JournalFile
*old_file
, *new_file
= NULL
;
3462 if (!old_file
->writable
)
3465 /* Is this a journal file that was passed to us as fd? If so, we synthesized a path name for it, and we refuse
3466 * rotation, since we don't know the actual path, and couldn't rename the file hence. */
3467 if (path_startswith(old_file
->path
, "/proc/self/fd"))
3470 if (!endswith(old_file
->path
, ".journal"))
3473 l
= strlen(old_file
->path
);
3474 r
= asprintf(&p
, "%.*s@" SD_ID128_FORMAT_STR
"-%016"PRIx64
"-%016"PRIx64
".journal",
3475 (int) l
- 8, old_file
->path
,
3476 SD_ID128_FORMAT_VAL(old_file
->header
->seqnum_id
),
3477 le64toh((*f
)->header
->head_entry_seqnum
),
3478 le64toh((*f
)->header
->head_entry_realtime
));
3482 /* Try to rename the file to the archived version. If the file
3483 * already was deleted, we'll get ENOENT, let's ignore that
3485 r
= rename(old_file
->path
, p
);
3486 if (r
< 0 && errno
!= ENOENT
)
3489 /* Sync the rename to disk */
3490 (void) fsync_directory_of_file(old_file
->fd
);
3492 /* Set as archive so offlining commits w/state=STATE_ARCHIVED.
3493 * Previously we would set old_file->header->state to STATE_ARCHIVED directly here,
3494 * but journal_file_set_offline() short-circuits when state != STATE_ONLINE, which
3495 * would result in the rotated journal never getting fsync() called before closing.
3496 * Now we simply queue the archive state by setting an archive bit, leaving the state
3497 * as STATE_ONLINE so proper offlining occurs. */
3498 old_file
->archive
= true;
3500 /* Currently, btrfs is not very good with out write patterns
3501 * and fragments heavily. Let's defrag our journal files when
3502 * we archive them */
3503 old_file
->defrag_on_close
= true;
3505 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
);
3507 if (deferred_closes
&&
3508 set_put(deferred_closes
, old_file
) >= 0)
3509 (void) journal_file_set_offline(old_file
, false);
3511 (void) journal_file_close(old_file
);
3517 int journal_file_open_reliably(
3523 JournalMetrics
*metrics
,
3524 MMapCache
*mmap_cache
,
3525 Set
*deferred_closes
,
3526 JournalFile
*template,
3527 JournalFile
**ret
) {
3531 _cleanup_free_
char *p
= NULL
;
3533 r
= journal_file_open(-1, fname
, flags
, mode
, compress
, seal
, metrics
, mmap_cache
, deferred_closes
, template, ret
);
3535 -EBADMSG
, /* Corrupted */
3536 -ENODATA
, /* Truncated */
3537 -EHOSTDOWN
, /* Other machine */
3538 -EPROTONOSUPPORT
, /* Incompatible feature */
3539 -EBUSY
, /* Unclean shutdown */
3540 -ESHUTDOWN
, /* Already archived */
3541 -EIO
, /* IO error, including SIGBUS on mmap */
3542 -EIDRM
, /* File has been deleted */
3543 -ETXTBSY
)) /* File is from the future */
3546 if ((flags
& O_ACCMODE
) == O_RDONLY
)
3549 if (!(flags
& O_CREAT
))
3552 if (!endswith(fname
, ".journal"))
3555 /* The file is corrupted. Rotate it away and try it again (but only once) */
3558 if (asprintf(&p
, "%.*s@%016"PRIx64
"-%016"PRIx64
".journal~",
3560 now(CLOCK_REALTIME
),
3564 if (rename(fname
, p
) < 0)
3567 /* btrfs doesn't cope well with our write pattern and
3568 * fragments heavily. Let's defrag all files we rotate */
3570 (void) chattr_path(p
, 0, FS_NOCOW_FL
);
3571 (void) btrfs_defrag(p
);
3573 log_warning_errno(r
, "File %s corrupted or uncleanly shut down, renaming and replacing.", fname
);
3575 return journal_file_open(-1, fname
, flags
, mode
, compress
, seal
, metrics
, mmap_cache
, deferred_closes
, template, ret
);
3578 int journal_file_copy_entry(JournalFile
*from
, JournalFile
*to
, Object
*o
, uint64_t p
, uint64_t *seqnum
, Object
**ret
, uint64_t *offset
) {
3580 uint64_t q
, xor_hash
= 0;
3593 ts
.monotonic
= le64toh(o
->entry
.monotonic
);
3594 ts
.realtime
= le64toh(o
->entry
.realtime
);
3596 n
= journal_file_entry_n_items(o
);
3597 /* alloca() can't take 0, hence let's allocate at least one */
3598 items
= alloca(sizeof(EntryItem
) * MAX(1u, n
));
3600 for (i
= 0; i
< n
; i
++) {
3607 q
= le64toh(o
->entry
.items
[i
].object_offset
);
3608 le_hash
= o
->entry
.items
[i
].hash
;
3610 r
= journal_file_move_to_object(from
, OBJECT_DATA
, q
, &o
);
3614 if (le_hash
!= o
->data
.hash
)
3617 l
= le64toh(o
->object
.size
) - offsetof(Object
, data
.payload
);
3620 /* We hit the limit on 32bit machines */
3621 if ((uint64_t) t
!= l
)
3624 if (o
->object
.flags
& OBJECT_COMPRESSION_MASK
) {
3625 #if HAVE_XZ || HAVE_LZ4
3628 r
= decompress_blob(o
->object
.flags
& OBJECT_COMPRESSION_MASK
,
3629 o
->data
.payload
, l
, &from
->compress_buffer
, &from
->compress_buffer_size
, &rsize
, 0);
3633 data
= from
->compress_buffer
;
3636 return -EPROTONOSUPPORT
;
3639 data
= o
->data
.payload
;
3641 r
= journal_file_append_data(to
, data
, l
, &u
, &h
);
3645 xor_hash
^= le64toh(u
->data
.hash
);
3646 items
[i
].object_offset
= htole64(h
);
3647 items
[i
].hash
= u
->data
.hash
;
3649 r
= journal_file_move_to_object(from
, OBJECT_ENTRY
, p
, &o
);
3654 r
= journal_file_append_entry_internal(to
, &ts
, xor_hash
, items
, n
, seqnum
, ret
, offset
);
3656 if (mmap_cache_got_sigbus(to
->mmap
, to
->cache_fd
))
3662 void journal_reset_metrics(JournalMetrics
*m
) {
3665 /* Set everything to "pick automatic values". */
3667 *m
= (JournalMetrics
) {
3668 .min_use
= (uint64_t) -1,
3669 .max_use
= (uint64_t) -1,
3670 .min_size
= (uint64_t) -1,
3671 .max_size
= (uint64_t) -1,
3672 .keep_free
= (uint64_t) -1,
3673 .n_max_files
= (uint64_t) -1,
3677 void journal_default_metrics(JournalMetrics
*m
, int fd
) {
3678 char a
[FORMAT_BYTES_MAX
], b
[FORMAT_BYTES_MAX
], c
[FORMAT_BYTES_MAX
], d
[FORMAT_BYTES_MAX
], e
[FORMAT_BYTES_MAX
];
3685 if (fstatvfs(fd
, &ss
) >= 0)
3686 fs_size
= ss
.f_frsize
* ss
.f_blocks
;
3688 log_debug_errno(errno
, "Failed to detremine disk size: %m");
3692 if (m
->max_use
== (uint64_t) -1) {
3695 m
->max_use
= PAGE_ALIGN(fs_size
/ 10); /* 10% of file system size */
3697 if (m
->max_use
> DEFAULT_MAX_USE_UPPER
)
3698 m
->max_use
= DEFAULT_MAX_USE_UPPER
;
3700 if (m
->max_use
< DEFAULT_MAX_USE_LOWER
)
3701 m
->max_use
= DEFAULT_MAX_USE_LOWER
;
3703 m
->max_use
= DEFAULT_MAX_USE_LOWER
;
3705 m
->max_use
= PAGE_ALIGN(m
->max_use
);
3707 if (m
->max_use
!= 0 && m
->max_use
< JOURNAL_FILE_SIZE_MIN
*2)
3708 m
->max_use
= JOURNAL_FILE_SIZE_MIN
*2;
3711 if (m
->min_use
== (uint64_t) -1)
3712 m
->min_use
= DEFAULT_MIN_USE
;
3714 if (m
->min_use
> m
->max_use
)
3715 m
->min_use
= m
->max_use
;
3717 if (m
->max_size
== (uint64_t) -1) {
3718 m
->max_size
= PAGE_ALIGN(m
->max_use
/ 8); /* 8 chunks */
3720 if (m
->max_size
> DEFAULT_MAX_SIZE_UPPER
)
3721 m
->max_size
= DEFAULT_MAX_SIZE_UPPER
;
3723 m
->max_size
= PAGE_ALIGN(m
->max_size
);
3725 if (m
->max_size
!= 0) {
3726 if (m
->max_size
< JOURNAL_FILE_SIZE_MIN
)
3727 m
->max_size
= JOURNAL_FILE_SIZE_MIN
;
3729 if (m
->max_use
!= 0 && m
->max_size
*2 > m
->max_use
)
3730 m
->max_use
= m
->max_size
*2;
3733 if (m
->min_size
== (uint64_t) -1)
3734 m
->min_size
= JOURNAL_FILE_SIZE_MIN
;
3736 m
->min_size
= PAGE_ALIGN(m
->min_size
);
3738 if (m
->min_size
< JOURNAL_FILE_SIZE_MIN
)
3739 m
->min_size
= JOURNAL_FILE_SIZE_MIN
;
3741 if (m
->max_size
!= 0 && m
->min_size
> m
->max_size
)
3742 m
->max_size
= m
->min_size
;
3745 if (m
->keep_free
== (uint64_t) -1) {
3748 m
->keep_free
= PAGE_ALIGN(fs_size
* 3 / 20); /* 15% of file system size */
3750 if (m
->keep_free
> DEFAULT_KEEP_FREE_UPPER
)
3751 m
->keep_free
= DEFAULT_KEEP_FREE_UPPER
;
3754 m
->keep_free
= DEFAULT_KEEP_FREE
;
3757 if (m
->n_max_files
== (uint64_t) -1)
3758 m
->n_max_files
= DEFAULT_N_MAX_FILES
;
3760 log_debug("Fixed min_use=%s max_use=%s max_size=%s min_size=%s keep_free=%s n_max_files=%" PRIu64
,
3761 format_bytes(a
, sizeof(a
), m
->min_use
),
3762 format_bytes(b
, sizeof(b
), m
->max_use
),
3763 format_bytes(c
, sizeof(c
), m
->max_size
),
3764 format_bytes(d
, sizeof(d
), m
->min_size
),
3765 format_bytes(e
, sizeof(e
), m
->keep_free
),
3769 int journal_file_get_cutoff_realtime_usec(JournalFile
*f
, usec_t
*from
, usec_t
*to
) {
3775 if (f
->header
->head_entry_realtime
== 0)
3778 *from
= le64toh(f
->header
->head_entry_realtime
);
3782 if (f
->header
->tail_entry_realtime
== 0)
3785 *to
= le64toh(f
->header
->tail_entry_realtime
);
3791 int journal_file_get_cutoff_monotonic_usec(JournalFile
*f
, sd_id128_t boot_id
, usec_t
*from
, usec_t
*to
) {
3799 r
= find_data_object_by_boot_id(f
, boot_id
, &o
, &p
);
3803 if (le64toh(o
->data
.n_entries
) <= 0)
3807 r
= journal_file_move_to_object(f
, OBJECT_ENTRY
, le64toh(o
->data
.entry_offset
), &o
);
3811 *from
= le64toh(o
->entry
.monotonic
);
3815 r
= journal_file_move_to_object(f
, OBJECT_DATA
, p
, &o
);
3819 r
= generic_array_get_plus_one(f
,
3820 le64toh(o
->data
.entry_offset
),
3821 le64toh(o
->data
.entry_array_offset
),
3822 le64toh(o
->data
.n_entries
)-1,
3827 *to
= le64toh(o
->entry
.monotonic
);
3833 bool journal_file_rotate_suggested(JournalFile
*f
, usec_t max_file_usec
) {
3837 /* If we gained new header fields we gained new features,
3838 * hence suggest a rotation */
3839 if (le64toh(f
->header
->header_size
) < sizeof(Header
)) {
3840 log_debug("%s uses an outdated header, suggesting rotation.", f
->path
);
3844 /* Let's check if the hash tables grew over a certain fill
3845 * level (75%, borrowing this value from Java's hash table
3846 * implementation), and if so suggest a rotation. To calculate
3847 * the fill level we need the n_data field, which only exists
3848 * in newer versions. */
3850 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
))
3851 if (le64toh(f
->header
->n_data
) * 4ULL > (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
3852 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.",
3854 100.0 * (double) le64toh(f
->header
->n_data
) / ((double) (le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
))),
3855 le64toh(f
->header
->n_data
),
3856 le64toh(f
->header
->data_hash_table_size
) / sizeof(HashItem
),
3857 (unsigned long long) f
->last_stat
.st_size
,
3858 f
->last_stat
.st_size
/ le64toh(f
->header
->n_data
));
3862 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
))
3863 if (le64toh(f
->header
->n_fields
) * 4ULL > (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
)) * 3ULL) {
3864 log_debug("Field hash table of %s has a fill level at %.1f (%"PRIu64
" of %"PRIu64
" items), suggesting rotation.",
3866 100.0 * (double) le64toh(f
->header
->n_fields
) / ((double) (le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
))),
3867 le64toh(f
->header
->n_fields
),
3868 le64toh(f
->header
->field_hash_table_size
) / sizeof(HashItem
));
3872 /* Are the data objects properly indexed by field objects? */
3873 if (JOURNAL_HEADER_CONTAINS(f
->header
, n_data
) &&
3874 JOURNAL_HEADER_CONTAINS(f
->header
, n_fields
) &&
3875 le64toh(f
->header
->n_data
) > 0 &&
3876 le64toh(f
->header
->n_fields
) == 0)
3879 if (max_file_usec
> 0) {
3882 h
= le64toh(f
->header
->head_entry_realtime
);
3883 t
= now(CLOCK_REALTIME
);
3885 if (h
> 0 && t
> h
+ max_file_usec
)