Google Git
Sign in
third-party-mirror / systemd / f229149d865bb087742add2415d0c7c321294591 / . / src / libsystemd / sd-journal / journal-file.c
blob: f622acbd5118f51bc73b5999c5dca40f1a097d49 [file] [log] [blame]
/* SPDX-License-Identifier: LGPL-2.1-or-later */
#include <errno.h>
#include <fcntl.h>
#include <linux/fs.h>
#include <linux/magic.h>
#include <pthread.h>
#include <stddef.h>
#include <sys/mman.h>
#include <sys/statvfs.h>
#include <sys/uio.h>
#include <unistd.h>
#include "sd-event.h"
#include "alloc-util.h"
#include "chattr-util.h"
#include "compress.h"
#include "env-util.h"
#include "fd-util.h"
#include "format-util.h"
#include "fs-util.h"
#include "id128-util.h"
#if HAVE_GCRYPT
#include "journal-authenticate.h"
#endif
#include "journal-def.h"
#include "journal-file.h"
#include "journal-internal.h"
#include "lookup3.h"
#include "memory-util.h"
#include "path-util.h"
#include "random-util.h"
#include "set.h"
#include "sort-util.h"
#include "stat-util.h"
#include "string-table.h"
#include "string-util.h"
#include "strv.h"
#include "sync-util.h"
#include "user-util.h"
#include "xattr-util.h"
#define DEFAULT_DATA_HASH_TABLE_SIZE (2047ULL*sizeof(HashItem))
#define DEFAULT_FIELD_HASH_TABLE_SIZE (333ULL*sizeof(HashItem))
#define DEFAULT_COMPRESS_THRESHOLD (512ULL)
#define MIN_COMPRESS_THRESHOLD (8ULL)
/* This is the minimum journal file size */
#define JOURNAL_FILE_SIZE_MIN (512 * 1024ULL) /* 512 KiB */
#define JOURNAL_COMPACT_SIZE_MAX UINT32_MAX /* 4 GiB */
/* These are the lower and upper bounds if we deduce the max_use value
* from the file system size */
#define MAX_USE_LOWER (1 * 1024 * 1024ULL) /* 1 MiB */
#define MAX_USE_UPPER (4 * 1024 * 1024 * 1024ULL) /* 4 GiB */
/* Those are the lower and upper bounds for the minimal use limit,
* i.e. how much we'll use even if keep_free suggests otherwise. */
#define MIN_USE_LOW (1 * 1024 * 1024ULL) /* 1 MiB */
#define MIN_USE_HIGH (16 * 1024 * 1024ULL) /* 16 MiB */
/* This is the upper bound if we deduce max_size from max_use */
#define MAX_SIZE_UPPER (128 * 1024 * 1024ULL) /* 128 MiB */
/* This is the upper bound if we deduce the keep_free value from the
* file system size */
#define KEEP_FREE_UPPER (4 * 1024 * 1024 * 1024ULL) /* 4 GiB */
/* This is the keep_free value when we can't determine the system
* size */
#define DEFAULT_KEEP_FREE (1024 * 1024ULL) /* 1 MB */
/* This is the default maximum number of journal files to keep around. */
#define DEFAULT_N_MAX_FILES 100
/* n_data was the first entry we added after the initial file format design */
#define HEADER_SIZE_MIN ALIGN64(offsetof(Header, n_data))
/* How many entries to keep in the entry array chain cache at max */
#define CHAIN_CACHE_MAX 20
/* How much to increase the journal file size at once each time we allocate something new. */
#define FILE_SIZE_INCREASE (8 * 1024 * 1024ULL) /* 8MB */
/* Reread fstat() of the file for detecting deletions at least this often */
#define LAST_STAT_REFRESH_USEC (5*USEC_PER_SEC)
/* The mmap context to use for the header we pick as one above the last defined typed */
#define CONTEXT_HEADER _OBJECT_TYPE_MAX
/* Longest hash chain to rotate after */
#define HASH_CHAIN_DEPTH_MAX 100
#ifdef __clang__
# pragma GCC diagnostic ignored "-Waddress-of-packed-member"
#endif
static int mmap_prot_from_open_flags(int flags) {
switch (flags & O_ACCMODE) {
case O_RDONLY:
return PROT_READ;
case O_WRONLY:
return PROT_WRITE;
case O_RDWR:
return PROT_READ|PROT_WRITE;
default:
assert_not_reached();
}
}
int journal_file_tail_end_by_pread(JournalFile *f, uint64_t *ret_offset) {
uint64_t p;
int r;
assert(f);
assert(f->header);
assert(ret_offset);
/* Same as journal_file_tail_end_by_mmap() below, but operates with pread() to avoid the mmap cache
* (and thus is thread safe) */
p = le64toh(f->header->tail_object_offset);
if (p == 0)
p = le64toh(f->header->header_size);
else {
Object tail;
uint64_t sz;
r = journal_file_read_object_header(f, OBJECT_UNUSED, p, &tail);
if (r < 0)
return r;
sz = le64toh(tail.object.size);
if (sz > UINT64_MAX - sizeof(uint64_t) + 1)
return -EBADMSG;
sz = ALIGN64(sz);
if (p > UINT64_MAX - sz)
return -EBADMSG;
p += sz;
}
*ret_offset = p;
return 0;
}
int journal_file_tail_end_by_mmap(JournalFile *f, uint64_t *ret_offset) {
uint64_t p;
int r;
assert(f);
assert(f->header);
assert(ret_offset);
/* Same as journal_file_tail_end_by_pread() above, but operates with the usual mmap logic */
p = le64toh(f->header->tail_object_offset);
if (p == 0)
p = le64toh(f->header->header_size);
else {
Object *tail;
uint64_t sz;
r = journal_file_move_to_object(f, OBJECT_UNUSED, p, &tail);
if (r < 0)
return r;
sz = le64toh(READ_NOW(tail->object.size));
if (sz > UINT64_MAX - sizeof(uint64_t) + 1)
return -EBADMSG;
sz = ALIGN64(sz);
if (p > UINT64_MAX - sz)
return -EBADMSG;
p += sz;
}
*ret_offset = p;
return 0;
}
int journal_file_set_offline_thread_join(JournalFile *f) {
int r;
assert(f);
if (f->offline_state == OFFLINE_JOINED)
return 0;
r = pthread_join(f->offline_thread, NULL);
if (r)
return -r;
f->offline_state = OFFLINE_JOINED;
if (mmap_cache_fd_got_sigbus(f->cache_fd))
return -EIO;
return 0;
}
static int journal_file_set_online(JournalFile *f) {
bool wait = true;
assert(f);
if (!journal_file_writable(f))
return -EPERM;
if (f->fd < 0 || !f->header)
return -EINVAL;
while (wait) {
switch (f->offline_state) {
case OFFLINE_JOINED:
/* No offline thread, no need to wait. */
wait = false;
break;
case OFFLINE_SYNCING: {
OfflineState tmp_state = OFFLINE_SYNCING;
if (!__atomic_compare_exchange_n(&f->offline_state, &tmp_state, OFFLINE_CANCEL,
false, __ATOMIC_SEQ_CST, __ATOMIC_SEQ_CST))
continue;
}
/* Canceled syncing prior to offlining, no need to wait. */
wait = false;
break;
case OFFLINE_AGAIN_FROM_SYNCING: {
OfflineState tmp_state = OFFLINE_AGAIN_FROM_SYNCING;
if (!__atomic_compare_exchange_n(&f->offline_state, &tmp_state, OFFLINE_CANCEL,
false, __ATOMIC_SEQ_CST, __ATOMIC_SEQ_CST))
continue;
}
/* Canceled restart from syncing, no need to wait. */
wait = false;
break;
case OFFLINE_AGAIN_FROM_OFFLINING: {
OfflineState tmp_state = OFFLINE_AGAIN_FROM_OFFLINING;
if (!__atomic_compare_exchange_n(&f->offline_state, &tmp_state, OFFLINE_CANCEL,
false, __ATOMIC_SEQ_CST, __ATOMIC_SEQ_CST))
continue;
}
/* Canceled restart from offlining, must wait for offlining to complete however. */
_fallthrough_;
default: {
int r;
r = journal_file_set_offline_thread_join(f);
if (r < 0)
return r;
wait = false;
break;
}
}
}
if (mmap_cache_fd_got_sigbus(f->cache_fd))
return -EIO;
switch (f->header->state) {
case STATE_ONLINE:
return 0;
case STATE_OFFLINE:
f->header->state = STATE_ONLINE;
(void) fsync(f->fd);
return 0;
default:
return -EINVAL;
}
}
JournalFile* journal_file_close(JournalFile *f) {
if (!f)
return NULL;
if (f->cache_fd)
mmap_cache_fd_free(f->cache_fd);
if (f->close_fd)
safe_close(f->fd);
free(f->path);
ordered_hashmap_free_free(f->chain_cache);
#if HAVE_COMPRESSION
free(f->compress_buffer);
#endif
#if HAVE_GCRYPT
if (f->fss_file)
munmap(f->fss_file, PAGE_ALIGN(f->fss_file_size));
else
free(f->fsprg_state);
free(f->fsprg_seed);
if (f->hmac)
gcry_md_close(f->hmac);
#endif
return mfree(f);
}
static bool keyed_hash_requested(void) {
int r;
r = getenv_bool("SYSTEMD_JOURNAL_KEYED_HASH");
if (r >= 0)
return r;
if (r != -ENXIO)
log_debug_errno(r, "Failed to parse $SYSTEMD_JOURNAL_KEYED_HASH environment variable, ignoring: %m");
return true;
}
static bool compact_mode_requested(void) {
int r;
r = getenv_bool("SYSTEMD_JOURNAL_COMPACT");
if (r >= 0)
return r;
if (r != -ENXIO)
log_debug_errno(r, "Failed to parse $SYSTEMD_JOURNAL_COMPACT environment variable, ignoring: %m");
return true;
}
static int journal_file_init_header(
JournalFile *f,
JournalFileFlags file_flags,
JournalFile *template) {
bool seal = false;
ssize_t k;
int r;
assert(f);
#if HAVE_GCRYPT
/* Try to load the FSPRG state, and if we can't, then just don't do sealing */
seal = FLAGS_SET(file_flags, JOURNAL_SEAL) && journal_file_fss_load(f) >= 0;
#endif
Header h = {
.header_size = htole64(ALIGN64(sizeof(h))),
.incompatible_flags = htole32(
FLAGS_SET(file_flags, JOURNAL_COMPRESS) * COMPRESSION_TO_HEADER_INCOMPATIBLE_FLAG(DEFAULT_COMPRESSION) |
keyed_hash_requested() * HEADER_INCOMPATIBLE_KEYED_HASH |
compact_mode_requested() * HEADER_INCOMPATIBLE_COMPACT),
.compatible_flags = htole32(seal * HEADER_COMPATIBLE_SEALED),
};
assert_cc(sizeof(h.signature) == sizeof(HEADER_SIGNATURE));
memcpy(h.signature, HEADER_SIGNATURE, sizeof(HEADER_SIGNATURE));
r = sd_id128_randomize(&h.file_id);
if (r < 0)
return r;
if (template) {
h.seqnum_id = template->header->seqnum_id;
h.tail_entry_seqnum = template->header->tail_entry_seqnum;
} else
h.seqnum_id = h.file_id;
k = pwrite(f->fd, &h, sizeof(h), 0);
if (k < 0)
return -errno;
if (k != sizeof(h))
return -EIO;
return 0;
}
static int journal_file_refresh_header(JournalFile *f) {
int r;
assert(f);
assert(f->header);
r = sd_id128_get_machine(&f->header->machine_id);
if (r < 0) {
if (!ERRNO_IS_MACHINE_ID_UNSET(r))
return r;
/* don't have a machine-id, let's continue without */
f->header->machine_id = SD_ID128_NULL;
}
r = sd_id128_get_boot(&f->header->boot_id);
if (r < 0)
return r;
r = journal_file_set_online(f);
/* Sync the online state to disk; likely just created a new file, also sync the directory this file
* is located in. */
(void) fsync_full(f->fd);
return r;
}
static bool warn_wrong_flags(const JournalFile *f, bool compatible) {
const uint32_t any = compatible ? HEADER_COMPATIBLE_ANY : HEADER_INCOMPATIBLE_ANY,
supported = compatible ? HEADER_COMPATIBLE_SUPPORTED : HEADER_INCOMPATIBLE_SUPPORTED;
const char *type = compatible ? "compatible" : "incompatible";
uint32_t flags;
assert(f);
assert(f->header);
flags = le32toh(compatible ? f->header->compatible_flags : f->header->incompatible_flags);
if (flags & ~supported) {
if (flags & ~any)
log_debug("Journal file %s has unknown %s flags 0x%"PRIx32,
f->path, type, flags & ~any);
flags = (flags & any) & ~supported;
if (flags) {
const char* strv[6];
size_t n = 0;
_cleanup_free_ char *t = NULL;
if (compatible) {
if (flags & HEADER_COMPATIBLE_SEALED)
strv[n++] = "sealed";
} else {
if (flags & HEADER_INCOMPATIBLE_COMPRESSED_XZ)
strv[n++] = "xz-compressed";
if (flags & HEADER_INCOMPATIBLE_COMPRESSED_LZ4)
strv[n++] = "lz4-compressed";
if (flags & HEADER_INCOMPATIBLE_COMPRESSED_ZSTD)
strv[n++] = "zstd-compressed";
if (flags & HEADER_INCOMPATIBLE_KEYED_HASH)
strv[n++] = "keyed-hash";
if (flags & HEADER_INCOMPATIBLE_COMPACT)
strv[n++] = "compact";
}
strv[n] = NULL;
assert(n < ELEMENTSOF(strv));
t = strv_join((char**) strv, ", ");
log_debug("Journal file %s uses %s %s %s disabled at compilation time.",
f->path, type, n > 1 ? "flags" : "flag", strnull(t));
}
return true;
}
return false;
}
static int journal_file_verify_header(JournalFile *f) {
uint64_t arena_size, header_size;
assert(f);
assert(f->header);
if (memcmp(f->header->signature, HEADER_SIGNATURE, 8))
return -EBADMSG;
/* In both read and write mode we refuse to open files with incompatible
* flags we don't know. */
if (warn_wrong_flags(f, false))
return -EPROTONOSUPPORT;
/* When open for writing we refuse to open files with compatible flags, too. */
if (journal_file_writable(f) && warn_wrong_flags(f, true))
return -EPROTONOSUPPORT;
if (f->header->state >= _STATE_MAX)
return -EBADMSG;
header_size = le64toh(READ_NOW(f->header->header_size));
/* The first addition was n_data, so check that we are at least this large */
if (header_size < HEADER_SIZE_MIN)
return -EBADMSG;
/* When open for writing we refuse to open files with a mismatch of the header size, i.e. writing to
* files implementing older or new header structures. */
if (journal_file_writable(f) && header_size != sizeof(Header))
return -EPROTONOSUPPORT;
if (JOURNAL_HEADER_SEALED(f->header) && !JOURNAL_HEADER_CONTAINS(f->header, n_entry_arrays))
return -EBADMSG;
arena_size = le64toh(READ_NOW(f->header->arena_size));
if (UINT64_MAX - header_size < arena_size || header_size + arena_size > (uint64_t) f->last_stat.st_size)
return -ENODATA;
if (le64toh(f->header->tail_object_offset) > header_size + arena_size)
return -ENODATA;
if (!VALID64(le64toh(f->header->data_hash_table_offset)) ||
!VALID64(le64toh(f->header->field_hash_table_offset)) ||
!VALID64(le64toh(f->header->tail_object_offset)) ||
!VALID64(le64toh(f->header->entry_array_offset)))
return -ENODATA;
if (journal_file_writable(f)) {
sd_id128_t machine_id;
uint8_t state;
int r;
r = sd_id128_get_machine(&machine_id);
if (r < 0)
return r;
if (!sd_id128_equal(machine_id, f->header->machine_id))
return log_debug_errno(SYNTHETIC_ERRNO(EHOSTDOWN),
"Trying to open journal file from different host for writing, refusing.");
state = f->header->state;
if (state == STATE_ARCHIVED)
return -ESHUTDOWN; /* Already archived */
if (state == STATE_ONLINE)
return log_debug_errno(SYNTHETIC_ERRNO(EBUSY),
"Journal file %s is already online. Assuming unclean closing.",
f->path);
if (state != STATE_OFFLINE)
return log_debug_errno(SYNTHETIC_ERRNO(EBUSY),
"Journal file %s has unknown state %i.",
f->path, state);
if (f->header->field_hash_table_size == 0 || f->header->data_hash_table_size == 0)
return -EBADMSG;
/* Don't permit appending to files from the future. Because otherwise the realtime timestamps wouldn't
* be strictly ordered in the entries in the file anymore, and we can't have that since it breaks
* bisection. */
if (le64toh(f->header->tail_entry_realtime) > now(CLOCK_REALTIME))
return log_debug_errno(SYNTHETIC_ERRNO(ETXTBSY),
"Journal file %s is from the future, refusing to append new data to it that'd be older.",
f->path);
}
return 0;
}
int journal_file_fstat(JournalFile *f) {
int r;
assert(f);
assert(f->fd >= 0);
if (fstat(f->fd, &f->last_stat) < 0)
return -errno;
f->last_stat_usec = now(CLOCK_MONOTONIC);
/* Refuse dealing with files that aren't regular */
r = stat_verify_regular(&f->last_stat);
if (r < 0)
return r;
/* Refuse appending to files that are already deleted */
if (f->last_stat.st_nlink <= 0)
return -EIDRM;
return 0;
}
static int journal_file_allocate(JournalFile *f, uint64_t offset, uint64_t size) {
uint64_t old_size, new_size, old_header_size, old_arena_size;
int r;
assert(f);
assert(f->header);
/* We assume that this file is not sparse, and we know that for sure, since we always call
* posix_fallocate() ourselves */
if (size > PAGE_ALIGN_DOWN(UINT64_MAX) - offset)
return -EINVAL;
if (mmap_cache_fd_got_sigbus(f->cache_fd))
return -EIO;
old_header_size = le64toh(READ_NOW(f->header->header_size));
old_arena_size = le64toh(READ_NOW(f->header->arena_size));
if (old_arena_size > PAGE_ALIGN_DOWN(UINT64_MAX) - old_header_size)
return -EBADMSG;
old_size = old_header_size + old_arena_size;
new_size = MAX(PAGE_ALIGN(offset + size), old_header_size);
if (new_size <= old_size) {
/* We already pre-allocated enough space, but before
* we write to it, let's check with fstat() if the
* file got deleted, in order make sure we don't throw
* away the data immediately. Don't check fstat() for
* all writes though, but only once ever 10s. */
if (f->last_stat_usec + LAST_STAT_REFRESH_USEC > now(CLOCK_MONOTONIC))
return 0;
return journal_file_fstat(f);
}
/* Allocate more space. */
if (f->metrics.max_size > 0 && new_size > f->metrics.max_size)
return -E2BIG;
/* Refuse to go over 4G in compact mode so offsets can be stored in 32-bit. */
if (JOURNAL_HEADER_COMPACT(f->header) && new_size > UINT32_MAX)
return -E2BIG;
if (new_size > f->metrics.min_size && f->metrics.keep_free > 0) {
struct statvfs svfs;
if (fstatvfs(f->fd, &svfs) >= 0) {
uint64_t available;
available = LESS_BY((uint64_t) svfs.f_bfree * (uint64_t) svfs.f_bsize, f->metrics.keep_free);
if (new_size - old_size > available)
return -E2BIG;
}
}
/* Increase by larger blocks at once */
new_size = DIV_ROUND_UP(new_size, FILE_SIZE_INCREASE) * FILE_SIZE_INCREASE;
if (f->metrics.max_size > 0 && new_size > f->metrics.max_size)
new_size = f->metrics.max_size;
/* Note that the glibc fallocate() fallback is very
inefficient, hence we try to minimize the allocation area
as we can. */
r = posix_fallocate_loop(f->fd, old_size, new_size - old_size);
if (r < 0)
return r;
f->header->arena_size = htole64(new_size - old_header_size);
return journal_file_fstat(f);
}
static unsigned type_to_context(ObjectType type) {
/* One context for each type, plus one catch-all for the rest */
assert_cc(_OBJECT_TYPE_MAX <= MMAP_CACHE_MAX_CONTEXTS);
assert_cc(CONTEXT_HEADER < MMAP_CACHE_MAX_CONTEXTS);
return type > OBJECT_UNUSED && type < _OBJECT_TYPE_MAX ? type : 0;
}
static int journal_file_move_to(
JournalFile *f,
ObjectType type,
bool keep_always,
uint64_t offset,
uint64_t size,
void **ret) {
int r;
assert(f);
assert(ret);
/* This function may clear, overwrite, or alter previously cached entries. After this function has
* been called, all objects except for one obtained by this function are invalidated and must be
* re-read before use. */
if (size <= 0)
return -EINVAL;
if (size > UINT64_MAX - offset)
return -EBADMSG;
/* Avoid SIGBUS on invalid accesses */
if (offset + size > (uint64_t) f->last_stat.st_size) {
/* Hmm, out of range? Let's refresh the fstat() data
* first, before we trust that check. */
r = journal_file_fstat(f);
if (r < 0)
return r;
if (offset + size > (uint64_t) f->last_stat.st_size)
return -EADDRNOTAVAIL;
}
return mmap_cache_fd_get(f->cache_fd, type_to_context(type), keep_always, offset, size, &f->last_stat, ret);
}
static uint64_t minimum_header_size(JournalFile *f, Object *o) {
static const uint64_t table[] = {
[OBJECT_DATA] = sizeof(DataObject),
[OBJECT_FIELD] = sizeof(FieldObject),
[OBJECT_ENTRY] = sizeof(EntryObject),
[OBJECT_DATA_HASH_TABLE] = sizeof(HashTableObject),
[OBJECT_FIELD_HASH_TABLE] = sizeof(HashTableObject),
[OBJECT_ENTRY_ARRAY] = sizeof(EntryArrayObject),
[OBJECT_TAG] = sizeof(TagObject),
};
assert(f);
assert(o);
if (o->object.type == OBJECT_DATA)
return journal_file_data_payload_offset(f);
if (o->object.type >= ELEMENTSOF(table) || table[o->object.type] <= 0)
return sizeof(ObjectHeader);
return table[o->object.type];
}
static int check_object_header(JournalFile *f, Object *o, ObjectType type, uint64_t offset) {
uint64_t s;
assert(f);
assert(o);
s = le64toh(READ_NOW(o->object.size));
if (s == 0)
return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
"Attempt to move to uninitialized object: %" PRIu64,
offset);
if (s < sizeof(ObjectHeader))
return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
"Attempt to move to overly short object: %" PRIu64,
offset);
if (o->object.type <= OBJECT_UNUSED)
return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
"Attempt to move to object with invalid type: %" PRIu64,
offset);
if (type > OBJECT_UNUSED && o->object.type != type)
return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
"Attempt to move to object of unexpected type: %" PRIu64,
offset);
if (s < minimum_header_size(f, o))
return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
"Attempt to move to truncated object: %" PRIu64,
offset);
return 0;
}
/* Lightweight object checks. We want this to be fast, so that we won't
* slowdown every journal_file_move_to_object() call too much. */
static int check_object(JournalFile *f, Object *o, uint64_t offset) {
assert(f);
assert(o);
switch (o->object.type) {
case OBJECT_DATA:
if ((le64toh(o->data.entry_offset) == 0) ^ (le64toh(o->data.n_entries) == 0))
return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
"Bad n_entries: %" PRIu64 ": %" PRIu64,
le64toh(o->data.n_entries),
offset);
if (le64toh(o->object.size) <= journal_file_data_payload_offset(f))
return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
"Bad object size (<= %zu): %" PRIu64 ": %" PRIu64,
journal_file_data_payload_offset(f),
le64toh(o->object.size),
offset);
if (!VALID64(le64toh(o->data.next_hash_offset)) ||
!VALID64(le64toh(o->data.next_field_offset)) ||
!VALID64(le64toh(o->data.entry_offset)) ||
!VALID64(le64toh(o->data.entry_array_offset)))
return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
"Invalid offset, next_hash_offset=" OFSfmt ", next_field_offset=" OFSfmt ", entry_offset=" OFSfmt ", entry_array_offset=" OFSfmt ": %" PRIu64,
le64toh(o->data.next_hash_offset),
le64toh(o->data.next_field_offset),
le64toh(o->data.entry_offset),
le64toh(o->data.entry_array_offset),
offset);
break;
case OBJECT_FIELD:
if (le64toh(o->object.size) <= offsetof(Object, field.payload))
return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
"Bad field size (<= %zu): %" PRIu64 ": %" PRIu64,
offsetof(Object, field.payload),
le64toh(o->object.size),
offset);
if (!VALID64(le64toh(o->field.next_hash_offset)) ||
!VALID64(le64toh(o->field.head_data_offset)))
return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
"Invalid offset, next_hash_offset=" OFSfmt ", head_data_offset=" OFSfmt ": %" PRIu64,
le64toh(o->field.next_hash_offset),
le64toh(o->field.head_data_offset),
offset);
break;
case OBJECT_ENTRY: {
uint64_t sz;
sz = le64toh(READ_NOW(o->object.size));
if (sz < offsetof(Object, entry.items) ||
(sz - offsetof(Object, entry.items)) % journal_file_entry_item_size(f) != 0)
return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
"Bad entry size (<= %zu): %" PRIu64 ": %" PRIu64,
offsetof(Object, entry.items),
sz,
offset);
if ((sz - offsetof(Object, entry.items)) / journal_file_entry_item_size(f) <= 0)
return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
"Invalid number items in entry: %" PRIu64 ": %" PRIu64,
(sz - offsetof(Object, entry.items)) / journal_file_entry_item_size(f),
offset);
if (le64toh(o->entry.seqnum) <= 0)
return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
"Invalid entry seqnum: %" PRIx64 ": %" PRIu64,
le64toh(o->entry.seqnum),
offset);
if (!VALID_REALTIME(le64toh(o->entry.realtime)))
return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
"Invalid entry realtime timestamp: %" PRIu64 ": %" PRIu64,
le64toh(o->entry.realtime),
offset);
if (!VALID_MONOTONIC(le64toh(o->entry.monotonic)))
return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
"Invalid entry monotonic timestamp: %" PRIu64 ": %" PRIu64,
le64toh(o->entry.monotonic),
offset);
break;
}
case OBJECT_DATA_HASH_TABLE:
case OBJECT_FIELD_HASH_TABLE: {
uint64_t sz;
sz = le64toh(READ_NOW(o->object.size));
if (sz < offsetof(Object, hash_table.items) ||
(sz - offsetof(Object, hash_table.items)) % sizeof(HashItem) != 0 ||
(sz - offsetof(Object, hash_table.items)) / sizeof(HashItem) <= 0)
return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
"Invalid %s hash table size: %" PRIu64 ": %" PRIu64,
o->object.type == OBJECT_DATA_HASH_TABLE ? "data" : "field",
sz,
offset);
break;
}
case OBJECT_ENTRY_ARRAY: {
uint64_t sz;
sz = le64toh(READ_NOW(o->object.size));
if (sz < offsetof(Object, entry_array.items) ||
(sz - offsetof(Object, entry_array.items)) % journal_file_entry_array_item_size(f) != 0 ||
(sz - offsetof(Object, entry_array.items)) / journal_file_entry_array_item_size(f) <= 0)
return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
"Invalid object entry array size: %" PRIu64 ": %" PRIu64,
sz,
offset);
if (!VALID64(le64toh(o->entry_array.next_entry_array_offset)))
return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
"Invalid object entry array next_entry_array_offset: " OFSfmt ": %" PRIu64,
le64toh(o->entry_array.next_entry_array_offset),
offset);
break;
}
case OBJECT_TAG:
if (le64toh(o->object.size) != sizeof(TagObject))
return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
"Invalid object tag size: %" PRIu64 ": %" PRIu64,
le64toh(o->object.size),
offset);
if (!VALID_EPOCH(le64toh(o->tag.epoch)))
return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
"Invalid object tag epoch: %" PRIu64 ": %" PRIu64,
le64toh(o->tag.epoch), offset);
break;
}
return 0;
}
int journal_file_move_to_object(JournalFile *f, ObjectType type, uint64_t offset, Object **ret) {
int r;
Object *o;
assert(f);
/* Even if this function fails, it may clear, overwrite, or alter previously cached entries. After
* this function has been called, all objects except for one obtained by this function are
* invalidated and must be re-read before use.. */
/* Objects may only be located at multiple of 64 bit */
if (!VALID64(offset))
return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
"Attempt to move to object at non-64bit boundary: %" PRIu64,
offset);
/* Object may not be located in the file header */
if (offset < le64toh(f->header->header_size))
return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
"Attempt to move to object located in file header: %" PRIu64,
offset);
r = journal_file_move_to(f, type, false, offset, sizeof(ObjectHeader), (void**) &o);
if (r < 0)
return r;
r = check_object_header(f, o, type, offset);
if (r < 0)
return r;
r = journal_file_move_to(f, type, false, offset, le64toh(READ_NOW(o->object.size)), (void**) &o);
if (r < 0)
return r;
r = check_object_header(f, o, type, offset);
if (r < 0)
return r;
r = check_object(f, o, offset);
if (r < 0)
return r;
if (ret)
*ret = o;
return 0;
}
int journal_file_read_object_header(JournalFile *f, ObjectType type, uint64_t offset, Object *ret) {
ssize_t n;
Object o;
int r;
assert(f);
/* Objects may only be located at multiple of 64 bit */
if (!VALID64(offset))
return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
"Attempt to read object at non-64bit boundary: %" PRIu64,
offset);
/* Object may not be located in the file header */
if (offset < le64toh(f->header->header_size))
return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
"Attempt to read object located in file header: %" PRIu64,
offset);
/* This will likely read too much data but it avoids having to call pread() twice. */
n = pread(f->fd, &o, sizeof(o), offset);
if (n < 0)
return log_debug_errno(errno, "Failed to read journal file at offset: %" PRIu64,
offset);
if ((size_t) n < sizeof(o.object))
return log_debug_errno(SYNTHETIC_ERRNO(EIO),
"Failed to read short object at offset: %" PRIu64,
offset);
r = check_object_header(f, &o, type, offset);
if (r < 0)
return r;
if ((size_t) n < minimum_header_size(f, &o))
return log_debug_errno(SYNTHETIC_ERRNO(EIO),
"Short read while reading object: %" PRIu64,
offset);
r = check_object(f, &o, offset);
if (r < 0)
return r;
if (ret)
*ret = o;
return 0;
}
static uint64_t inc_seqnum(uint64_t seqnum) {
if (seqnum < UINT64_MAX-1)
return seqnum + 1;
return 1; /* skip over UINT64_MAX and 0 when we run out of seqnums and start again */
}
static uint64_t journal_file_entry_seqnum(
JournalFile *f,
uint64_t *seqnum) {
uint64_t next_seqnum;
assert(f);
assert(f->header);
/* Picks a new sequence number for the entry we are about to add and returns it. */
next_seqnum = inc_seqnum(le64toh(f->header->tail_entry_seqnum));
/* If an external seqnum counter was passed, we update both the local and the external one, and set
* it to the maximum of both */
if (seqnum)
*seqnum = next_seqnum = MAX(inc_seqnum(*seqnum), next_seqnum);
f->header->tail_entry_seqnum = htole64(next_seqnum);
if (f->header->head_entry_seqnum == 0)
f->header->head_entry_seqnum = htole64(next_seqnum);
return next_seqnum;
}
int journal_file_append_object(
JournalFile *f,
ObjectType type,
uint64_t size,
Object **ret_object,
uint64_t *ret_offset) {
int r;
uint64_t p;
Object *o;
assert(f);
assert(f->header);
assert(type > OBJECT_UNUSED && type < _OBJECT_TYPE_MAX);
assert(size >= sizeof(ObjectHeader));
r = journal_file_set_online(f);
if (r < 0)
return r;
r = journal_file_tail_end_by_mmap(f, &p);
if (r < 0)
return r;
r = journal_file_allocate(f, p, size);
if (r < 0)
return r;
r = journal_file_move_to(f, type, false, p, size, (void**) &o);
if (r < 0)
return r;
o->object = (ObjectHeader) {
.type = type,
.size = htole64(size),
};
f->header->tail_object_offset = htole64(p);
f->header->n_objects = htole64(le64toh(f->header->n_objects) + 1);
if (ret_object)
*ret_object = o;
if (ret_offset)
*ret_offset = p;
return 0;
}
static int journal_file_setup_data_hash_table(JournalFile *f) {
uint64_t s, p;
Object *o;
int r;
assert(f);
assert(f->header);
/* We estimate that we need 1 hash table entry per 768 bytes
of journal file and we want to make sure we never get
beyond 75% fill level. Calculate the hash table size for
the maximum file size based on these metrics. */
s = (f->metrics.max_size * 4 / 768 / 3) * sizeof(HashItem);
if (s < DEFAULT_DATA_HASH_TABLE_SIZE)
s = DEFAULT_DATA_HASH_TABLE_SIZE;
log_debug("Reserving %"PRIu64" entries in data hash table.", s / sizeof(HashItem));
r = journal_file_append_object(f,
OBJECT_DATA_HASH_TABLE,
offsetof(Object, hash_table.items) + s,
&o, &p);
if (r < 0)
return r;
memzero(o->hash_table.items, s);
f->header->data_hash_table_offset = htole64(p + offsetof(Object, hash_table.items));
f->header->data_hash_table_size = htole64(s);
return 0;
}
static int journal_file_setup_field_hash_table(JournalFile *f) {
uint64_t s, p;
Object *o;
int r;
assert(f);
assert(f->header);
/* We use a fixed size hash table for the fields as this
* number should grow very slowly only */
s = DEFAULT_FIELD_HASH_TABLE_SIZE;
log_debug("Reserving %"PRIu64" entries in field hash table.", s / sizeof(HashItem));
r = journal_file_append_object(f,
OBJECT_FIELD_HASH_TABLE,
offsetof(Object, hash_table.items) + s,
&o, &p);
if (r < 0)
return r;
memzero(o->hash_table.items, s);
f->header->field_hash_table_offset = htole64(p + offsetof(Object, hash_table.items));
f->header->field_hash_table_size = htole64(s);
return 0;
}
int journal_file_map_data_hash_table(JournalFile *f) {
uint64_t s, p;
void *t;
int r;
assert(f);
assert(f->header);
if (f->data_hash_table)
return 0;
p = le64toh(f->header->data_hash_table_offset);
s = le64toh(f->header->data_hash_table_size);
r = journal_file_move_to(f,
OBJECT_DATA_HASH_TABLE,
true,
p, s,
&t);
if (r < 0)
return r;
f->data_hash_table = t;
return 0;
}
int journal_file_map_field_hash_table(JournalFile *f) {
uint64_t s, p;
void *t;
int r;
assert(f);
assert(f->header);
if (f->field_hash_table)
return 0;
p = le64toh(f->header->field_hash_table_offset);
s = le64toh(f->header->field_hash_table_size);
r = journal_file_move_to(f,
OBJECT_FIELD_HASH_TABLE,
true,
p, s,
&t);
if (r < 0)
return r;
f->field_hash_table = t;
return 0;
}
static int journal_file_link_field(
JournalFile *f,
Object *o,
uint64_t offset,
uint64_t hash) {
uint64_t p, h, m;
int r;
assert(f);
assert(f->header);
assert(f->field_hash_table);
assert(o);
assert(offset > 0);
if (o->object.type != OBJECT_FIELD)
return -EINVAL;
m = le64toh(READ_NOW(f->header->field_hash_table_size)) / sizeof(HashItem);
if (m <= 0)
return -EBADMSG;
/* This might alter the window we are looking at */
o->field.next_hash_offset = o->field.head_data_offset = 0;
h = hash % m;
p = le64toh(f->field_hash_table[h].tail_hash_offset);
if (p == 0)
f->field_hash_table[h].head_hash_offset = htole64(offset);
else {
r = journal_file_move_to_object(f, OBJECT_FIELD, p, &o);
if (r < 0)
return r;
o->field.next_hash_offset = htole64(offset);
}
f->field_hash_table[h].tail_hash_offset = htole64(offset);
if (JOURNAL_HEADER_CONTAINS(f->header, n_fields))
f->header->n_fields = htole64(le64toh(f->header->n_fields) + 1);
return 0;
}
static int journal_file_link_data(
JournalFile *f,
Object *o,
uint64_t offset,
uint64_t hash) {
uint64_t p, h, m;
int r;
assert(f);
assert(f->header);
assert(f->data_hash_table);
assert(o);
assert(offset > 0);
if (o->object.type != OBJECT_DATA)
return -EINVAL;
m = le64toh(READ_NOW(f->header->data_hash_table_size)) / sizeof(HashItem);
if (m <= 0)
return -EBADMSG;
/* This might alter the window we are looking at */
o->data.next_hash_offset = o->data.next_field_offset = 0;
o->data.entry_offset = o->data.entry_array_offset = 0;
o->data.n_entries = 0;
h = hash % m;
p = le64toh(f->data_hash_table[h].tail_hash_offset);
if (p == 0)
/* Only entry in the hash table is easy */
f->data_hash_table[h].head_hash_offset = htole64(offset);
else {
/* Move back to the previous data object, to patch in
* pointer */
r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
if (r < 0)
return r;
o->data.next_hash_offset = htole64(offset);
}
f->data_hash_table[h].tail_hash_offset = htole64(offset);
if (JOURNAL_HEADER_CONTAINS(f->header, n_data))
f->header->n_data = htole64(le64toh(f->header->n_data) + 1);
return 0;
}
static int get_next_hash_offset(
JournalFile *f,
uint64_t *p,
le64_t *next_hash_offset,
uint64_t *depth,
le64_t *header_max_depth) {
uint64_t nextp;
assert(f);
assert(p);
assert(next_hash_offset);
assert(depth);
nextp = le64toh(READ_NOW(*next_hash_offset));
if (nextp > 0) {
if (nextp <= *p) /* Refuse going in loops */
return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
"Detected hash item loop in %s, refusing.", f->path);
(*depth)++;
/* If the depth of this hash chain is larger than all others we have seen so far, record it */
if (header_max_depth && journal_file_writable(f))
*header_max_depth = htole64(MAX(*depth, le64toh(*header_max_depth)));
}
*p = nextp;
return 0;
}
int journal_file_find_field_object_with_hash(
JournalFile *f,
const void *field,
uint64_t size,
uint64_t hash,
Object **ret_object,
uint64_t *ret_offset) {
uint64_t p, osize, h, m, depth = 0;
int r;
assert(f);
assert(f->header);
assert(field);
assert(size > 0);
/* If the field hash table is empty, we can't find anything */
if (le64toh(f->header->field_hash_table_size) <= 0)
return 0;
/* Map the field hash table, if it isn't mapped yet. */
r = journal_file_map_field_hash_table(f);
if (r < 0)
return r;
osize = offsetof(Object, field.payload) + size;
m = le64toh(READ_NOW(f->header->field_hash_table_size)) / sizeof(HashItem);
if (m <= 0)
return -EBADMSG;
h = hash % m;
p = le64toh(f->field_hash_table[h].head_hash_offset);
while (p > 0) {
Object *o;
r = journal_file_move_to_object(f, OBJECT_FIELD, p, &o);
if (r < 0)
return r;
if (le64toh(o->field.hash) == hash &&
le64toh(o->object.size) == osize &&
memcmp(o->field.payload, field, size) == 0) {
if (ret_object)
*ret_object = o;
if (ret_offset)
*ret_offset = p;
return 1;
}
r = get_next_hash_offset(
f,
&p,
&o->field.next_hash_offset,
&depth,
JOURNAL_HEADER_CONTAINS(f->header, field_hash_chain_depth) ? &f->header->field_hash_chain_depth : NULL);
if (r < 0)
return r;
}
return 0;
}
uint64_t journal_file_hash_data(
JournalFile *f,
const void *data,
size_t sz) {
assert(f);
assert(f->header);
assert(data || sz == 0);
/* We try to unify our codebase on siphash, hence new-styled journal files utilizing the keyed hash
* function use siphash. Old journal files use the Jenkins hash. */
if (JOURNAL_HEADER_KEYED_HASH(f->header))
return siphash24(data, sz, f->header->file_id.bytes);
return jenkins_hash64(data, sz);
}
int journal_file_find_field_object(
JournalFile *f,
const void *field,
uint64_t size,
Object **ret_object,
uint64_t *ret_offset) {
assert(f);
assert(field);
assert(size > 0);
return journal_file_find_field_object_with_hash(
f,
field, size,
journal_file_hash_data(f, field, size),
ret_object, ret_offset);
}
int journal_file_find_data_object_with_hash(
JournalFile *f,
const void *data,
uint64_t size,
uint64_t hash,
Object **ret_object,
uint64_t *ret_offset) {
uint64_t p, h, m, depth = 0;
int r;
assert(f);
assert(f->header);
assert(data || size == 0);
/* If there's no data hash table, then there's no entry. */
if (le64toh(f->header->data_hash_table_size) <= 0)
return 0;
/* Map the data hash table, if it isn't mapped yet. */
r = journal_file_map_data_hash_table(f);
if (r < 0)
return r;
m = le64toh(READ_NOW(f->header->data_hash_table_size)) / sizeof(HashItem);
if (m <= 0)
return -EBADMSG;
h = hash % m;
p = le64toh(f->data_hash_table[h].head_hash_offset);
while (p > 0) {
Object *o;
void *d;
size_t rsize;
r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
if (r < 0)
return r;
if (le64toh(o->data.hash) != hash)
goto next;
r = journal_file_data_payload(f, o, p, NULL, 0, 0, &d, &rsize);
if (r < 0)
return r;
assert(r > 0); /* journal_file_data_payload() always returns > 0 if no field is provided. */
if (memcmp_nn(data, size, d, rsize) == 0) {
if (ret_object)
*ret_object = o;
if (ret_offset)
*ret_offset = p;
return 1;
}
next:
r = get_next_hash_offset(
f,
&p,
&o->data.next_hash_offset,
&depth,
JOURNAL_HEADER_CONTAINS(f->header, data_hash_chain_depth) ? &f->header->data_hash_chain_depth : NULL);
if (r < 0)
return r;
}
return 0;
}
int journal_file_find_data_object(
JournalFile *f,
const void *data,
uint64_t size,
Object **ret_object,
uint64_t *ret_offset) {
assert(f);
assert(data || size == 0);
return journal_file_find_data_object_with_hash(
f,
data, size,
journal_file_hash_data(f, data, size),
ret_object, ret_offset);
}
bool journal_field_valid(const char *p, size_t l, bool allow_protected) {
/* We kinda enforce POSIX syntax recommendations for
environment variables here, but make a couple of additional
requirements.
http://pubs.opengroup.org/onlinepubs/000095399/basedefs/xbd_chap08.html */
assert(p);
if (l == SIZE_MAX)
l = strlen(p);
/* No empty field names */
if (l <= 0)
return false;
/* Don't allow names longer than 64 chars */
if (l > 64)
return false;
/* Variables starting with an underscore are protected */
if (!allow_protected && p[0] == '_')
return false;
/* Don't allow digits as first character */
if (ascii_isdigit(p[0]))
return false;
/* Only allow A-Z0-9 and '_' */
for (const char *a = p; a < p + l; a++)
if ((*a < 'A' || *a > 'Z') &&
!ascii_isdigit(*a) &&
*a != '_')
return false;
return true;
}
static int journal_file_append_field(
JournalFile *f,
const void *field,
uint64_t size,
Object **ret_object,
uint64_t *ret_offset) {
uint64_t hash, p;
uint64_t osize;
Object *o;
int r;
assert(f);
assert(field);
assert(size > 0);
if (!journal_field_valid(field, size, true))
return -EBADMSG;
hash = journal_file_hash_data(f, field, size);
r = journal_file_find_field_object_with_hash(f, field, size, hash, ret_object, ret_offset);
if (r < 0)
return r;
if (r > 0)
return 0;
osize = offsetof(Object, field.payload) + size;
r = journal_file_append_object(f, OBJECT_FIELD, osize, &o, &p);
if (r < 0)
return r;
o->field.hash = htole64(hash);
memcpy(o->field.payload, field, size);
r = journal_file_link_field(f, o, p, hash);
if (r < 0)
return r;
/* The linking might have altered the window, so let's only pass the offset to hmac which will
* move to the object again if needed. */
#if HAVE_GCRYPT
r = journal_file_hmac_put_object(f, OBJECT_FIELD, NULL, p);
if (r < 0)
return r;
#endif
if (ret_object) {
r = journal_file_move_to_object(f, OBJECT_FIELD, p, ret_object);
if (r < 0)
return r;
}
if (ret_offset)
*ret_offset = p;
return 0;
}
static Compression maybe_compress_payload(JournalFile *f, uint8_t *dst, const uint8_t *src, uint64_t size, size_t *rsize) {
Compression compression = COMPRESSION_NONE;
assert(f);
assert(f->header);
#if HAVE_COMPRESSION
if (JOURNAL_FILE_COMPRESS(f) && size >= f->compress_threshold_bytes) {
compression = compress_blob(src, size, dst, size - 1, rsize);
if (compression > 0)
log_debug("Compressed data object %"PRIu64" -> %zu using %s",
size, *rsize, compression_to_string(compression));
else
/* Compression didn't work, we don't really care why, let's continue without compression */
compression = COMPRESSION_NONE;
}
#endif
return compression;
}
static int journal_file_append_data(
JournalFile *f,
const void *data,
uint64_t size,
Object **ret_object,
uint64_t *ret_offset) {
uint64_t hash, p, osize;
Object *o, *fo;
size_t rsize = 0;
Compression c;
const void *eq;
int r;
assert(f);
if (!data || size == 0)
return -EINVAL;
hash = journal_file_hash_data(f, data, size);
r = journal_file_find_data_object_with_hash(f, data, size, hash, ret_object, ret_offset);
if (r < 0)
return r;
if (r > 0)
return 0;
eq = memchr(data, '=', size);
if (!eq)
return -EINVAL;
osize = journal_file_data_payload_offset(f) + size;
r = journal_file_append_object(f, OBJECT_DATA, osize, &o, &p);
if (r < 0)
return r;
o->data.hash = htole64(hash);
c = maybe_compress_payload(f, journal_file_data_payload_field(f, o), data, size, &rsize);
if (c != COMPRESSION_NONE) {
o->object.size = htole64(journal_file_data_payload_offset(f) + rsize);
o->object.flags |= COMPRESSION_TO_OBJECT_FLAG(c);
} else
memcpy_safe(journal_file_data_payload_field(f, o), data, size);
r = journal_file_link_data(f, o, p, hash);
if (r < 0)
return r;
/* The linking might have altered the window, so let's refresh our pointer. */
r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
if (r < 0)
return r;
#if HAVE_GCRYPT
r = journal_file_hmac_put_object(f, OBJECT_DATA, o, p);
if (r < 0)
return r;
#endif
/* Create field object ... */
r = journal_file_append_field(f, data, (uint8_t*) eq - (uint8_t*) data, &fo, NULL);
if (r < 0)
return r;
/* ... and link it in. */
o->data.next_field_offset = fo->field.head_data_offset;
fo->field.head_data_offset = le64toh(p);
if (ret_object)
*ret_object = o;
if (ret_offset)
*ret_offset = p;
return 0;
}
static int maybe_decompress_payload(
JournalFile *f,
uint8_t *payload,
uint64_t size,
Compression compression,
const char *field,
size_t field_length,
size_t data_threshold,
void **ret_data,
size_t *ret_size) {
assert(f);
/* We can't read objects larger than 4G on a 32bit machine */
if ((uint64_t) (size_t) size != size)
return -E2BIG;
if (compression != COMPRESSION_NONE) {
#if HAVE_COMPRESSION
size_t rsize;
int r;
if (field) {
r = decompress_startswith(compression, payload, size, &f->compress_buffer, field,
field_length, '=');
if (r < 0)
return log_debug_errno(r,
"Cannot decompress %s object of length %" PRIu64 ": %m",
compression_to_string(compression),
size);
if (r == 0) {
if (ret_data)
*ret_data = NULL;
if (ret_size)
*ret_size = 0;
return 0;
}
}
r = decompress_blob(compression, payload, size, &f->compress_buffer, &rsize, 0);
if (r < 0)
return r;
if (ret_data)
*ret_data = f->compress_buffer;
if (ret_size)
*ret_size = rsize;
#else
return -EPROTONOSUPPORT;
#endif
} else {
if (field && (size < field_length + 1 || memcmp(payload, field, field_length) != 0 || payload[field_length] != '=')) {
if (ret_data)
*ret_data = NULL;
if (ret_size)
*ret_size = 0;
return 0;
}
if (ret_data)
*ret_data = payload;
if (ret_size)
*ret_size = (size_t) size;
}
return 1;
}
int journal_file_data_payload(
JournalFile *f,
Object *o,
uint64_t offset,
const char *field,
size_t field_length,
size_t data_threshold,
void **ret_data,
size_t *ret_size) {
uint64_t size;
Compression c;
int r;
assert(f);
assert(!field == (field_length == 0)); /* These must be specified together. */
if (!o) {
r = journal_file_move_to_object(f, OBJECT_DATA, offset, &o);
if (r < 0)
return r;
}
size = le64toh(READ_NOW(o->object.size));
if (size < journal_file_data_payload_offset(f))
return -EBADMSG;
size -= journal_file_data_payload_offset(f);
c = COMPRESSION_FROM_OBJECT(o);
if (c < 0)
return -EPROTONOSUPPORT;
return maybe_decompress_payload(f, journal_file_data_payload_field(f, o), size, c, field,
field_length, data_threshold, ret_data, ret_size);
}
uint64_t journal_file_entry_n_items(JournalFile *f, Object *o) {
uint64_t sz;
assert(f);
assert(o);
if (o->object.type != OBJECT_ENTRY)
return 0;
sz = le64toh(READ_NOW(o->object.size));
if (sz < offsetof(Object, entry.items))
return 0;
return (sz - offsetof(Object, entry.items)) / journal_file_entry_item_size(f);
}
uint64_t journal_file_entry_array_n_items(JournalFile *f, Object *o) {
uint64_t sz;
assert(f);
assert(o);
if (o->object.type != OBJECT_ENTRY_ARRAY)
return 0;
sz = le64toh(READ_NOW(o->object.size));
if (sz < offsetof(Object, entry_array.items))
return 0;
return (sz - offsetof(Object, entry_array.items)) / journal_file_entry_array_item_size(f);
}
uint64_t journal_file_hash_table_n_items(Object *o) {
uint64_t sz;
assert(o);
if (!IN_SET(o->object.type, OBJECT_DATA_HASH_TABLE, OBJECT_FIELD_HASH_TABLE))
return 0;
sz = le64toh(READ_NOW(o->object.size));
if (sz < offsetof(Object, hash_table.items))
return 0;
return (sz - offsetof(Object, hash_table.items)) / sizeof(HashItem);
}
static void write_entry_array_item(JournalFile *f, Object *o, uint64_t i, uint64_t p) {
assert(f);
assert(o);
if (JOURNAL_HEADER_COMPACT(f->header)) {
assert(p <= UINT32_MAX);
o->entry_array.items.compact[i] = htole32(p);
} else
o->entry_array.items.regular[i] = htole64(p);
}
static int link_entry_into_array(
JournalFile *f,
le64_t *first,
le64_t *idx,
le32_t *tail,
le32_t *tidx,
uint64_t p) {
uint64_t n = 0, ap = 0, q, i, a, hidx;
Object *o;
int r;
assert(f);
assert(f->header);
assert(first);
assert(idx);
assert(p > 0);
a = tail ? le32toh(*tail) : le64toh(*first);
hidx = le64toh(READ_NOW(*idx));
i = tidx ? le32toh(READ_NOW(*tidx)) : hidx;
while (a > 0) {
r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &o);
if (r < 0)
return r;
n = journal_file_entry_array_n_items(f, o);
if (i < n) {
write_entry_array_item(f, o, i, p);
*idx = htole64(hidx + 1);
if (tidx)
*tidx = htole32(le32toh(*tidx) + 1);
return 0;
}
i -= n;
ap = a;
a = le64toh(o->entry_array.next_entry_array_offset);
}
if (hidx > n)
n = (hidx+1) * 2;
else
n = n * 2;
if (n < 4)
n = 4;
r = journal_file_append_object(f, OBJECT_ENTRY_ARRAY,
offsetof(Object, entry_array.items) + n * journal_file_entry_array_item_size(f),
&o, &q);
if (r < 0)
return r;
#if HAVE_GCRYPT
r = journal_file_hmac_put_object(f, OBJECT_ENTRY_ARRAY, o, q);
if (r < 0)
return r;
#endif
write_entry_array_item(f, o, i, p);
if (ap == 0)
*first = htole64(q);
else {
r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, ap, &o);
if (r < 0)
return r;
o->entry_array.next_entry_array_offset = htole64(q);
}
if (tail)
*tail = htole32(q);
if (JOURNAL_HEADER_CONTAINS(f->header, n_entry_arrays))
f->header->n_entry_arrays = htole64(le64toh(f->header->n_entry_arrays) + 1);
*idx = htole64(hidx + 1);
if (tidx)
*tidx = htole32(1);
return 0;
}
static int link_entry_into_array_plus_one(
JournalFile *f,
le64_t *extra,
le64_t *first,
le64_t *idx,
le32_t *tail,
le32_t *tidx,
uint64_t p) {
uint64_t hidx;
int r;
assert(f);
assert(extra);
assert(first);
assert(idx);
assert(p > 0);
hidx = le64toh(READ_NOW(*idx));
if (hidx == UINT64_MAX)
return -EBADMSG;
if (hidx == 0)
*extra = htole64(p);
else {
le64_t i;
i = htole64(hidx - 1);
r = link_entry_into_array(f, first, &i, tail, tidx, p);
if (r < 0)
return r;
}
*idx = htole64(hidx + 1);
return 0;
}
static int journal_file_link_entry_item(JournalFile *f, uint64_t offset, uint64_t p) {
Object *o;
int r;
assert(f);
assert(offset > 0);
r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
if (r < 0)
return r;
return link_entry_into_array_plus_one(f,
&o->data.entry_offset,
&o->data.entry_array_offset,
&o->data.n_entries,
JOURNAL_HEADER_COMPACT(f->header) ? &o->data.compact.tail_entry_array_offset : NULL,
JOURNAL_HEADER_COMPACT(f->header) ? &o->data.compact.tail_entry_array_n_entries : NULL,
offset);
}
static int journal_file_link_entry(
JournalFile *f,
Object *o,
uint64_t offset,
const EntryItem items[],
size_t n_items) {
int r;
assert(f);
assert(f->header);
assert(o);
assert(offset > 0);
if (o->object.type != OBJECT_ENTRY)
return -EINVAL;
__atomic_thread_fence(__ATOMIC_SEQ_CST);
/* Link up the entry itself */
r = link_entry_into_array(f,
&f->header->entry_array_offset,
&f->header->n_entries,
JOURNAL_HEADER_CONTAINS(f->header, tail_entry_array_offset) ? &f->header->tail_entry_array_offset : NULL,
JOURNAL_HEADER_CONTAINS(f->header, tail_entry_array_n_entries) ? &f->header->tail_entry_array_n_entries : NULL,
offset);
if (r < 0)
return r;
/* log_debug("=> %s seqnr=%"PRIu64" n_entries=%"PRIu64, f->path, o->entry.seqnum, f->header->n_entries); */
if (f->header->head_entry_realtime == 0)
f->header->head_entry_realtime = o->entry.realtime;
f->header->tail_entry_realtime = o->entry.realtime;
f->header->tail_entry_monotonic = o->entry.monotonic;
/* Link up the items */
for (uint64_t i = 0; i < n_items; i++) {
int k;
/* If we fail to link an entry item because we can't allocate a new entry array, don't fail
* immediately but try to link the other entry items since it might still be possible to link
* those if they don't require a new entry array to be allocated. */
k = journal_file_link_entry_item(f, offset, items[i].object_offset);
if (k == -E2BIG)
r = k;
else if (k < 0)
return k;
}
return r;
}
static void write_entry_item(JournalFile *f, Object *o, uint64_t i, const EntryItem *item) {
assert(f);
assert(o);
assert(item);
if (JOURNAL_HEADER_COMPACT(f->header)) {
assert(item->object_offset <= UINT32_MAX);
o->entry.items.compact[i].object_offset = htole32(item->object_offset);
} else {
o->entry.items.regular[i].object_offset = htole64(item->object_offset);
o->entry.items.regular[i].hash = htole64(item->hash);
}
}
static int journal_file_append_entry_internal(
JournalFile *f,
const dual_timestamp *ts,
const sd_id128_t *boot_id,
uint64_t xor_hash,
const EntryItem items[],
size_t n_items,
uint64_t *seqnum,
Object **ret_object,
uint64_t *ret_offset) {
uint64_t np;
uint64_t osize;
Object *o;
int r;
assert(f);
assert(f->header);
assert(ts);
assert(items || n_items == 0);
if (f->strict_order) {
/* If requested be stricter with ordering in this journal file, to make searching via
* bisection fully deterministic. This is an optional feature, so that if desired journal
* files can be written where the ordering is not strictly enforced (in which case bisection
* will yield *a* result, but not the *only* result, when searching for points in
* time). Strict ordering mode is enabled when journald originally writes the files, but
* might not necessarily be if other tools (the remoting tools for example) write journal
* files from combined sources.
*
* Typically, if any of the errors generated here are seen journald will just rotate the
* journal files and start anew. */
if (ts->realtime < le64toh(f->header->tail_entry_realtime))
return log_debug_errno(SYNTHETIC_ERRNO(EREMCHG),
"Realtime timestamp %" PRIu64 " smaller than previous realtime "
"timestamp %" PRIu64 ", refusing entry.",
ts->realtime, le64toh(f->header->tail_entry_realtime));
if (!sd_id128_is_null(f->header->boot_id) && boot_id) {
if (!sd_id128_equal(f->header->boot_id, *boot_id))
return log_debug_errno(SYNTHETIC_ERRNO(EREMOTE),
"Boot ID to write is different from previous boot id, refusing entry.");
if (ts->monotonic < le64toh(f->header->tail_entry_monotonic))
return log_debug_errno(SYNTHETIC_ERRNO(ENOTNAM),
"Monotonic timestamp %" PRIu64 " smaller than previous monotonic "
"timestamp %" PRIu64 ", refusing entry.",
ts->monotonic, le64toh(f->header->tail_entry_monotonic));
}
}
osize = offsetof(Object, entry.items) + (n_items * journal_file_entry_item_size(f));
r = journal_file_append_object(f, OBJECT_ENTRY, osize, &o, &np);
if (r < 0)
return r;
o->entry.seqnum = htole64(journal_file_entry_seqnum(f, seqnum));
o->entry.realtime = htole64(ts->realtime);
o->entry.monotonic = htole64(ts->monotonic);
o->entry.xor_hash = htole64(xor_hash);
if (boot_id)
f->header->boot_id = *boot_id;
o->entry.boot_id = f->header->boot_id;
for (size_t i = 0; i < n_items; i++)
write_entry_item(f, o, i, &items[i]);
#if HAVE_GCRYPT
r = journal_file_hmac_put_object(f, OBJECT_ENTRY, o, np);
if (r < 0)
return r;
#endif
r = journal_file_link_entry(f, o, np, items, n_items);
if (r < 0)
return r;
if (ret_object)
*ret_object = o;
if (ret_offset)
*ret_offset = np;
return r;
}
void journal_file_post_change(JournalFile *f) {
assert(f);
if (f->fd < 0)
return;
/* inotify() does not receive IN_MODIFY events from file
* accesses done via mmap(). After each access we hence
* trigger IN_MODIFY by truncating the journal file to its
* current size which triggers IN_MODIFY. */
__atomic_thread_fence(__ATOMIC_SEQ_CST);
if (ftruncate(f->fd, f->last_stat.st_size) < 0)
log_debug_errno(errno, "Failed to truncate file to its own size: %m");
}
static int post_change_thunk(sd_event_source *timer, uint64_t usec, void *userdata) {
assert(userdata);
journal_file_post_change(userdata);
return 1;
}
static void schedule_post_change(JournalFile *f) {
sd_event *e;
int r;
assert(f);
assert(f->post_change_timer);
assert_se(e = sd_event_source_get_event(f->post_change_timer));
/* If we are already going down, post the change immediately. */
if (IN_SET(sd_event_get_state(e), SD_EVENT_EXITING, SD_EVENT_FINISHED))
goto fail;
r = sd_event_source_get_enabled(f->post_change_timer, NULL);
if (r < 0) {
log_debug_errno(r, "Failed to get ftruncate timer state: %m");
goto fail;
}
if (r > 0)
return;
r = sd_event_source_set_time_relative(f->post_change_timer, f->post_change_timer_period);
if (r < 0) {
log_debug_errno(r, "Failed to set time for scheduling ftruncate: %m");
goto fail;
}
r = sd_event_source_set_enabled(f->post_change_timer, SD_EVENT_ONESHOT);
if (r < 0) {
log_debug_errno(r, "Failed to enable scheduled ftruncate: %m");
goto fail;
}
return;
fail:
/* On failure, let's simply post the change immediately. */
journal_file_post_change(f);
}
/* Enable coalesced change posting in a timer on the provided sd_event instance */
int journal_file_enable_post_change_timer(JournalFile *f, sd_event *e, usec_t t) {
_cleanup_(sd_event_source_unrefp) sd_event_source *timer = NULL;
int r;
assert(f);
assert_return(!f->post_change_timer, -EINVAL);
assert(e);
assert(t);
r = sd_event_add_time(e, &timer, CLOCK_MONOTONIC, 0, 0, post_change_thunk, f);
if (r < 0)
return r;
r = sd_event_source_set_enabled(timer, SD_EVENT_OFF);
if (r < 0)
return r;
f->post_change_timer = TAKE_PTR(timer);
f->post_change_timer_period = t;
return r;
}
static int entry_item_cmp(const EntryItem *a, const EntryItem *b) {
return CMP(ASSERT_PTR(a)->object_offset, ASSERT_PTR(b)->object_offset);
}
static size_t remove_duplicate_entry_items(EntryItem items[], size_t n) {
size_t j = 1;
assert(items || n == 0);
if (n <= 1)
return n;
for (size_t i = 1; i < n; i++)
if (items[i].object_offset != items[j - 1].object_offset)
items[j++] = items[i];
return j;
}
int journal_file_append_entry(
JournalFile *f,
const dual_timestamp *ts,
const sd_id128_t *boot_id,
const struct iovec iovec[],
size_t n_iovec,
uint64_t *seqnum,
Object **ret_object,
uint64_t *ret_offset) {
_cleanup_free_ EntryItem *items_alloc = NULL;
EntryItem *items;
uint64_t xor_hash = 0;
struct dual_timestamp _ts;
sd_id128_t _boot_id;
int r;
assert(f);
assert(f->header);
assert(iovec);
assert(n_iovec > 0);
if (ts) {
if (!VALID_REALTIME(ts->realtime))
return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
"Invalid realtime timestamp %" PRIu64 ", refusing entry.",
ts->realtime);
if (!VALID_MONOTONIC(ts->monotonic))
return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
"Invalid monotomic timestamp %" PRIu64 ", refusing entry.",
ts->monotonic);
} else {
dual_timestamp_get(&_ts);
ts = &_ts;
}
if (!boot_id) {
r = sd_id128_get_boot(&_boot_id);
if (r < 0)
return r;
boot_id = &_boot_id;
}
#if HAVE_GCRYPT
r = journal_file_maybe_append_tag(f, ts->realtime);
if (r < 0)
return r;
#endif
if (n_iovec < ALLOCA_MAX / sizeof(EntryItem) / 2)
items = newa(EntryItem, n_iovec);
else {
items_alloc = new(EntryItem, n_iovec);
if (!items_alloc)
return -ENOMEM;
items = items_alloc;
}
for (size_t i = 0; i < n_iovec; i++) {
uint64_t p;
Object *o;
r = journal_file_append_data(f, iovec[i].iov_base, iovec[i].iov_len, &o, &p);
if (r < 0)
return r;
/* When calculating the XOR hash field, we need to take special care if the "keyed-hash"
* journal file flag is on. We use the XOR hash field to quickly determine the identity of a
* specific record, and give records with otherwise identical position (i.e. match in seqno,
* timestamp, …) a stable ordering. But for that we can't have it that the hash of the
* objects in each file is different since they are keyed. Hence let's calculate the Jenkins
* hash here for that. This also has the benefit that cursors for old and new journal files
* are completely identical (they include the XOR hash after all). For classic Jenkins-hash
* files things are easier, we can just take the value from the stored record directly. */
if (JOURNAL_HEADER_KEYED_HASH(f->header))
xor_hash ^= jenkins_hash64(iovec[i].iov_base, iovec[i].iov_len);
else
xor_hash ^= le64toh(o->data.hash);
items[i] = (EntryItem) {
.object_offset = p,
.hash = le64toh(o->data.hash),
};
}
/* Order by the position on disk, in order to improve seek
* times for rotating media. */
typesafe_qsort(items, n_iovec, entry_item_cmp);
n_iovec = remove_duplicate_entry_items(items, n_iovec);
r = journal_file_append_entry_internal(f, ts, boot_id, xor_hash, items, n_iovec, seqnum, ret_object, ret_offset);
/* If the memory mapping triggered a SIGBUS then we return an
* IO error and ignore the error code passed down to us, since
* it is very likely just an effect of a nullified replacement
* mapping page */
if (mmap_cache_fd_got_sigbus(f->cache_fd))
r = -EIO;
if (f->post_change_timer)
schedule_post_change(f);
else
journal_file_post_change(f);
return r;
}
typedef struct ChainCacheItem {
uint64_t first; /* the array at the beginning of the chain */
uint64_t array; /* the cached array */
uint64_t begin; /* the first item in the cached array */
uint64_t total; /* the total number of items in all arrays before this one in the chain */
uint64_t last_index; /* the last index we looked at, to optimize locality when bisecting */
} ChainCacheItem;
static void chain_cache_put(
OrderedHashmap *h,
ChainCacheItem *ci,
uint64_t first,
uint64_t array,
uint64_t begin,
uint64_t total,
uint64_t last_index) {
assert(h);
if (!ci) {
/* If the chain item to cache for this chain is the
* first one it's not worth caching anything */
if (array == first)
return;
if (ordered_hashmap_size(h) >= CHAIN_CACHE_MAX) {
ci = ordered_hashmap_steal_first(h);
assert(ci);
} else {
ci = new(ChainCacheItem, 1);
if (!ci)
return;
}
ci->first = first;
if (ordered_hashmap_put(h, &ci->first, ci) < 0) {
free(ci);
return;
}
} else
assert(ci->first == first);
ci->array = array;
ci->begin = begin;
ci->total = total;
ci->last_index = last_index;
}
static int bump_array_index(uint64_t *i, direction_t direction, uint64_t n) {
assert(i);
/* Increase or decrease the specified index, in the right direction. */
if (direction == DIRECTION_DOWN) {
if (*i >= n - 1)
return 0;
(*i)++;
} else {
if (*i <= 0)
return 0;
(*i)--;
}
return 1;
}
static int bump_entry_array(
JournalFile *f,
Object *o,
uint64_t offset,
uint64_t first,
direction_t direction,
uint64_t *ret) {
uint64_t p, q = 0;
int r;
assert(f);
assert(offset);
assert(ret);
if (direction == DIRECTION_DOWN) {
assert(o);
*ret = le64toh(o->entry_array.next_entry_array_offset);
return 0;
}
/* Entry array chains are a singly linked list, so to find the previous array in the chain, we have
* to start iterating from the top. */
p = first;
while (p > 0 && p != offset) {
r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, p, &o);
if (r < 0)
return r;
q = p;
p = le64toh(o->entry_array.next_entry_array_offset);
}
/* If we can't find the previous entry array in the entry array chain, we're likely dealing with a
* corrupted journal file. */
if (p == 0)
return -EBADMSG;
*ret = q;
return 0;
}
static int generic_array_get(
JournalFile *f,
uint64_t first,
uint64_t i,
direction_t direction,
Object **ret_object,
uint64_t *ret_offset) {
uint64_t p = 0, a, t = 0, k;
ChainCacheItem *ci;
Object *o;
int r;
assert(f);
/* FIXME: fix return value assignment on success. */
a = first;
/* Try the chain cache first */
ci = ordered_hashmap_get(f->chain_cache, &first);
if (ci && i > ci->total) {
a = ci->array;
i -= ci->total;
t = ci->total;
}
while (a > 0) {
r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &o);
if (IN_SET(r, -EBADMSG, -EADDRNOTAVAIL)) {
/* If there's corruption and we're going downwards, let's pretend we reached the
* final entry in the entry array chain. */
if (direction == DIRECTION_DOWN)
return 0;
/* If there's corruption and we're going upwards, move back to the previous entry
* array and start iterating entries from there. */
r = bump_entry_array(f, NULL, a, first, DIRECTION_UP, &a);
if (r < 0)
return r;
i = UINT64_MAX;
break;
}
if (r < 0)
return r;
k = journal_file_entry_array_n_items(f, o);
if (i < k)
break;
i -= k;
t += k;
a = le64toh(o->entry_array.next_entry_array_offset);
}
/* If we've found the right location, now look for the first non-corrupt entry object (in the right
* direction). */
while (a > 0) {
/* In the first iteration of the while loop, we reuse i, k and o from the previous while
* loop. */
if (i == UINT64_MAX) {
r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &o);
if (r < 0)
return r;
k = journal_file_entry_array_n_items(f, o);
if (k == 0)
break;
i = direction == DIRECTION_DOWN ? 0 : k - 1;
}
do {
p = journal_file_entry_array_item(f, o, i);
r = journal_file_move_to_object(f, OBJECT_ENTRY, p, ret_object);
if (r >= 0) {
/* Let's cache this item for the next invocation */
chain_cache_put(f->chain_cache, ci, first, a, journal_file_entry_array_item(f, o, 0), t, i);
if (ret_offset)
*ret_offset = p;
return 1;
}
if (!IN_SET(r, -EADDRNOTAVAIL, -EBADMSG))
return r;
/* OK, so this entry is borked. Most likely some entry didn't get synced to
* disk properly, let's see if the next one might work for us instead. */
log_debug_errno(r, "Entry item %" PRIu64 " is bad, skipping over it.", i);
} while (bump_array_index(&i, direction, k) > 0);
r = bump_entry_array(f, o, a, first, direction, &a);
if (r < 0)
return r;
t += k;
i = UINT64_MAX;
}
return 0;
}
static int generic_array_get_plus_one(
JournalFile *f,
uint64_t extra,
uint64_t first,
uint64_t i,
direction_t direction,
Object **ret_object,
uint64_t *ret_offset) {
int r;
assert(f);
/* FIXME: fix return value assignment on success. */
if (i == 0) {
r = journal_file_move_to_object(f, OBJECT_ENTRY, extra, ret_object);
if (IN_SET(r, -EADDRNOTAVAIL, -EBADMSG))
return generic_array_get(f, first, 0, direction, ret_object, ret_offset);
if (r < 0)
return r;
if (ret_offset)
*ret_offset = extra;
return 1;
}
return generic_array_get(f, first, i - 1, direction, ret_object, ret_offset);
}
enum {
TEST_FOUND,
TEST_LEFT,
TEST_RIGHT
};
static int generic_array_bisect(
JournalFile *f,
uint64_t first,
uint64_t n,
uint64_t needle,
int (*test_object)(JournalFile *f, uint64_t p, uint64_t needle),
direction_t direction,
Object **ret_object,
uint64_t *ret_offset,
uint64_t *ret_idx) {
/* Given an entry array chain, this function finds the object "closest" to the given needle in the
* chain, taking into account the provided direction. A function can be provided to determine how
* an object is matched against the given needle.
*
* Given a journal file, the offset of an object and the needle, the test_object() function should
* return TEST_LEFT if the needle is located earlier in the entry array chain, TEST_RIGHT if the
* needle is located later in the entry array chain and TEST_FOUND if the object matches the needle.
* If test_object() returns TEST_FOUND for a specific object, that object's information will be used
* to populate the return values of this function. If test_object() never returns TEST_FOUND, the
* return values are populated with the details of one of the objects closest to the needle. If the
* direction is DIRECTION_UP, the earlier object is used. Otherwise, the later object is used.
*/
uint64_t a, p, t = 0, i = 0, last_p = 0, last_index = UINT64_MAX;
bool subtract_one = false;
Object *array = NULL;
ChainCacheItem *ci;
int r;
assert(f);
assert(test_object);
/* Start with the first array in the chain */
a = first;
ci = ordered_hashmap_get(f->chain_cache, &first);
if (ci && n > ci->total && ci->begin != 0) {
/* Ah, we have iterated this bisection array chain
* previously! Let's see if we can skip ahead in the
* chain, as far as the last time. But we can't jump
* backwards in the chain, so let's check that
* first. */
r = test_object(f, ci->begin, needle);
if (r < 0)
return r;
if (r == TEST_LEFT) {
/* OK, what we are looking for is right of the
* begin of this EntryArray, so let's jump
* straight to previously cached array in the
* chain */
a = ci->array;
n -= ci->total;
t = ci->total;
last_index = ci->last_index;
}
}
while (a > 0) {
uint64_t left, right, k, lp;
r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &array);
if (r < 0)
return r;
k = journal_file_entry_array_n_items(f, array);
right = MIN(k, n);
if (right <= 0)
return 0;
i = right - 1;
lp = p = journal_file_entry_array_item(f, array, i);
if (p <= 0)
r = -EBADMSG;
else
r = test_object(f, p, needle);
if (r == -EBADMSG) {
log_debug_errno(r, "Encountered invalid entry while bisecting, cutting algorithm short. (1)");
n = i;
continue;
}
if (r < 0)
return r;
if (r == TEST_FOUND)
r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
if (r == TEST_RIGHT) {
left = 0;
right -= 1;
if (last_index != UINT64_MAX) {
assert(last_index <= right);
/* If we cached the last index we
* looked at, let's try to not to jump
* too wildly around and see if we can
* limit the range to look at early to
* the immediate neighbors of the last
* index we looked at. */
if (last_index > 0) {
uint64_t x = last_index - 1;
p = journal_file_entry_array_item(f, array, x);
if (p <= 0)
return -EBADMSG;
r = test_object(f, p, needle);
if (r < 0)
return r;
if (r == TEST_FOUND)
r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
if (r == TEST_RIGHT)
right = x;
else
left = x + 1;
}
if (last_index < right) {
uint64_t y = last_index + 1;
p = journal_file_entry_array_item(f, array, y);
if (p <= 0)
return -EBADMSG;
r = test_object(f, p, needle);
if (r < 0)
return r;
if (r == TEST_FOUND)
r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
if (r == TEST_RIGHT)
right = y;
else
left = y + 1;
}
}
for (;;) {
if (left == right) {
if (direction == DIRECTION_UP)
subtract_one = true;
i = left;
goto found;
}
assert(left < right);
i = (left + right) / 2;
p = journal_file_entry_array_item(f, array, i);
if (p <= 0)
r = -EBADMSG;
else
r = test_object(f, p, needle);
if (r == -EBADMSG) {
log_debug_errno(r, "Encountered invalid entry while bisecting, cutting algorithm short. (2)");
right = n = i;
continue;
}
if (r < 0)
return r;
if (r == TEST_FOUND)
r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
if (r == TEST_RIGHT)
right = i;
else
left = i + 1;
}
}
if (k >= n) {
if (direction == DIRECTION_UP) {
i = n;
subtract_one = true;
goto found;
}
return 0;
}
last_p = lp;
n -= k;
t += k;
last_index = UINT64_MAX;
a = le64toh(array->entry_array.next_entry_array_offset);
}
return 0;
found:
if (subtract_one && t == 0 && i == 0)
return 0;
/* Let's cache this item for the next invocation */
chain_cache_put(f->chain_cache, ci, first, a, journal_file_entry_array_item(f, array, 0), t, subtract_one ? (i > 0 ? i-1 : UINT64_MAX) : i);
if (subtract_one && i == 0)
p = last_p;
else if (subtract_one)
p = journal_file_entry_array_item(f, array, i - 1);
else
p = journal_file_entry_array_item(f, array, i);
if (ret_object) {
r = journal_file_move_to_object(f, OBJECT_ENTRY, p, ret_object);
if (r < 0)
return r;
}
if (ret_offset)
*ret_offset = p;
if (ret_idx)
*ret_idx = t + i + (subtract_one ? -1 : 0);
return 1;
}
static int generic_array_bisect_plus_one(
JournalFile *f,
uint64_t extra,
uint64_t first,
uint64_t n,
uint64_t needle,
int (*test_object)(JournalFile *f, uint64_t p, uint64_t needle),
direction_t direction,
Object **ret_object,
uint64_t *ret_offset,
uint64_t *ret_idx) {
int r;
bool step_back = false;
assert(f);
assert(test_object);
if (n <= 0)
return 0;
/* This bisects the array in object 'first', but first checks
* an extra */
r = test_object(f, extra, needle);
if (r < 0)
return r;
if (r == TEST_FOUND)
r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
/* if we are looking with DIRECTION_UP then we need to first
see if in the actual array there is a matching entry, and
return the last one of that. But if there isn't any we need
to return this one. Hence remember this, and return it
below. */
if (r == TEST_LEFT)
step_back = direction == DIRECTION_UP;
if (r == TEST_RIGHT) {
if (direction == DIRECTION_DOWN)
goto found;
else
return 0;
}
r = generic_array_bisect(f, first, n-1, needle, test_object, direction, ret_object, ret_offset, ret_idx);
if (r == 0 && step_back)
goto found;
if (r > 0 && ret_idx)
(*ret_idx)++;
return r;
found:
if (ret_object) {
r = journal_file_move_to_object(f, OBJECT_ENTRY, extra, ret_object);
if (r < 0)
return r;
}
if (ret_offset)
*ret_offset = extra;
if (ret_idx)
*ret_idx = 0;
return 1;
}
_pure_ static int test_object_offset(JournalFile *f, uint64_t p, uint64_t needle) {
assert(f);
assert(p > 0);
if (p == needle)
return TEST_FOUND;
else if (p < needle)
return TEST_LEFT;
else
return TEST_RIGHT;
}
int journal_file_move_to_entry_by_offset(
JournalFile *f,
uint64_t p,
direction_t direction,
Object **ret_object,
uint64_t *ret_offset) {
assert(f);
assert(f->header);
return generic_array_bisect(
f,
le64toh(f->header->entry_array_offset),
le64toh(f->header->n_entries),
p,
test_object_offset,
direction,
ret_object, ret_offset, NULL);
}
static int test_object_seqnum(JournalFile *f, uint64_t p, uint64_t needle) {
uint64_t sq;
Object *o;
int r;
assert(f);
assert(p > 0);
r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
if (r < 0)
return r;
sq = le64toh(READ_NOW(o->entry.seqnum));
if (sq == needle)
return TEST_FOUND;
else if (sq < needle)
return TEST_LEFT;
else
return TEST_RIGHT;
}
int journal_file_move_to_entry_by_seqnum(
JournalFile *f,
uint64_t seqnum,
direction_t direction,
Object **ret_object,
uint64_t *ret_offset) {
assert(f);
assert(f->header);
return generic_array_bisect(
f,
le64toh(f->header->entry_array_offset),
le64toh(f->header->n_entries),
seqnum,
test_object_seqnum,
direction,
ret_object, ret_offset, NULL);
}
static int test_object_realtime(JournalFile *f, uint64_t p, uint64_t needle) {
Object *o;
uint64_t rt;
int r;
assert(f);
assert(p > 0);
r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
if (r < 0)
return r;
rt = le64toh(READ_NOW(o->entry.realtime));
if (rt == needle)
return TEST_FOUND;
else if (rt < needle)
return TEST_LEFT;
else
return TEST_RIGHT;
}
int journal_file_move_to_entry_by_realtime(
JournalFile *f,
uint64_t realtime,
direction_t direction,
Object **ret_object,
uint64_t *ret_offset) {
assert(f);
assert(f->header);
return generic_array_bisect(
f,
le64toh(f->header->entry_array_offset),
le64toh(f->header->n_entries),
realtime,
test_object_realtime,
direction,
ret_object, ret_offset, NULL);
}
static int test_object_monotonic(JournalFile *f, uint64_t p, uint64_t needle) {
Object *o;
uint64_t m;
int r;
assert(f);
assert(p > 0);
r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
if (r < 0)
return r;
m = le64toh(READ_NOW(o->entry.monotonic));
if (m == needle)
return TEST_FOUND;
else if (m < needle)
return TEST_LEFT;
else
return TEST_RIGHT;
}
static int find_data_object_by_boot_id(
JournalFile *f,
sd_id128_t boot_id,
Object **ret_object,
uint64_t *ret_offset) {
char t[STRLEN("_BOOT_ID=") + 32 + 1] = "_BOOT_ID=";
assert(f);
sd_id128_to_string(boot_id, t + 9);
return journal_file_find_data_object(f, t, sizeof(t) - 1, ret_object, ret_offset);
}
int journal_file_move_to_entry_by_monotonic(
JournalFile *f,
sd_id128_t boot_id,
uint64_t monotonic,
direction_t direction,
Object **ret_object,
uint64_t *ret_offset) {
Object *o;
int r;
assert(f);
r = find_data_object_by_boot_id(f, boot_id, &o, NULL);
if (r < 0)
return r;
if (r == 0)
return -ENOENT;
return generic_array_bisect_plus_one(
f,
le64toh(o->data.entry_offset),
le64toh(o->data.entry_array_offset),
le64toh(o->data.n_entries),
monotonic,
test_object_monotonic,
direction,
ret_object, ret_offset, NULL);
}
void journal_file_reset_location(JournalFile *f) {
assert(f);
f->location_type = LOCATION_HEAD;
f->current_offset = 0;
f->current_seqnum = 0;
f->current_realtime = 0;
f->current_monotonic = 0;
zero(f->current_boot_id);
f->current_xor_hash = 0;
}
void journal_file_save_location(JournalFile *f, Object *o, uint64_t offset) {
assert(f);
assert(o);
f->location_type = LOCATION_SEEK;
f->current_offset = offset;
f->current_seqnum = le64toh(o->entry.seqnum);
f->current_realtime = le64toh(o->entry.realtime);
f->current_monotonic = le64toh(o->entry.monotonic);
f->current_boot_id = o->entry.boot_id;
f->current_xor_hash = le64toh(o->entry.xor_hash);
}
int journal_file_compare_locations(JournalFile *af, JournalFile *bf) {
int r;
assert(af);
assert(af->header);
assert(bf);
assert(bf->header);
assert(af->location_type == LOCATION_SEEK);
assert(bf->location_type == LOCATION_SEEK);
/* If contents, timestamps and seqnum match, these entries are
* identical. */
if (sd_id128_equal(af->current_boot_id, bf->current_boot_id) &&
af->current_monotonic == bf->current_monotonic &&
af->current_realtime == bf->current_realtime &&
af->current_xor_hash == bf->current_xor_hash &&
sd_id128_equal(af->header->seqnum_id, bf->header->seqnum_id) &&
af->current_seqnum == bf->current_seqnum)
return 0;
if (sd_id128_equal(af->header->seqnum_id, bf->header->seqnum_id)) {
/* If this is from the same seqnum source, compare
* seqnums */
r = CMP(af->current_seqnum, bf->current_seqnum);
if (r != 0)
return r;
/* Wow! This is weird, different data but the same
* seqnums? Something is borked, but let's make the
* best of it and compare by time. */
}
if (sd_id128_equal(af->current_boot_id, bf->current_boot_id)) {
/* If the boot id matches, compare monotonic time */
r = CMP(af->current_monotonic, bf->current_monotonic);
if (r != 0)
return r;
}
/* Otherwise, compare UTC time */
r = CMP(af->current_realtime, bf->current_realtime);
if (r != 0)
return r;
/* Finally, compare by contents */
return CMP(af->current_xor_hash, bf->current_xor_hash);
}
static bool check_properly_ordered(uint64_t new_offset, uint64_t old_offset, direction_t direction) {
/* Consider it an error if any of the two offsets is uninitialized */
if (old_offset == 0 || new_offset == 0)
return false;
/* If we go down, the new offset must be larger than the old one. */
return direction == DIRECTION_DOWN ?
new_offset > old_offset :
new_offset < old_offset;
}
int journal_file_next_entry(
JournalFile *f,
uint64_t p,
direction_t direction,
Object **ret_object,
uint64_t *ret_offset) {
uint64_t i, n, ofs;
int r;
assert(f);
assert(f->header);
/* FIXME: fix return value assignment. */
n = le64toh(READ_NOW(f->header->n_entries));
if (n <= 0)
return 0;
if (p == 0)
i = direction == DIRECTION_DOWN ? 0 : n - 1;
else {
r = generic_array_bisect(f,
le64toh(f->header->entry_array_offset),
le64toh(f->header->n_entries),
p,
test_object_offset,
DIRECTION_DOWN,
NULL, NULL,
&i);
if (r <= 0)
return r;
r = bump_array_index(&i, direction, n);
if (r <= 0)
return r;
}
/* And jump to it */
r = generic_array_get(f, le64toh(f->header->entry_array_offset), i, direction, ret_object, &ofs);
if (r <= 0)
return r;
/* Ensure our array is properly ordered. */
if (p > 0 && !check_properly_ordered(ofs, p, direction))
return log_debug_errno(SYNTHETIC_ERRNO(EBADMSG),
"%s: entry array not properly ordered at entry %" PRIu64,
f->path, i);
if (ret_offset)
*ret_offset = ofs;
return 1;
}
int journal_file_next_entry_for_data(
JournalFile *f,
Object *d,
direction_t direction,
Object **ret_object,
uint64_t *ret_offset) {
uint64_t i, n, ofs;
int r;
assert(f);
assert(d);
assert(d->object.type == OBJECT_DATA);
/* FIXME: fix return value assignment. */
n = le64toh(READ_NOW(d->data.n_entries));
if (n <= 0)
return n;
i = direction == DIRECTION_DOWN ? 0 : n - 1;
r = generic_array_get_plus_one(f,
le64toh(d->data.entry_offset),
le64toh(d->data.entry_array_offset),
i,
direction,
ret_object, &ofs);
if (r <= 0)
return r;
if (ret_offset)
*ret_offset = ofs;
return 1;
}
int journal_file_move_to_entry_by_offset_for_data(
JournalFile *f,
Object *d,
uint64_t p,
direction_t direction,
Object **ret, uint64_t *ret_offset) {
assert(f);
assert(d);
assert(d->object.type == OBJECT_DATA);
return generic_array_bisect_plus_one(
f,
le64toh(d->data.entry_offset),
le64toh(d->data.entry_array_offset),
le64toh(d->data.n_entries),
p,
test_object_offset,
direction,
ret, ret_offset, NULL);
}
int journal_file_move_to_entry_by_monotonic_for_data(
JournalFile *f,
Object *d,
sd_id128_t boot_id,
uint64_t monotonic,
direction_t direction,
Object **ret_object,
uint64_t *ret_offset) {
uint64_t b, z, entry_offset, entry_array_offset, n_entries;
Object *o;
int r;
assert(f);
assert(d);
assert(d->object.type == OBJECT_DATA);
/* Save all the required data before the data object gets invalidated. */
entry_offset = le64toh(READ_NOW(d->data.entry_offset));
entry_array_offset = le64toh(READ_NOW(d->data.entry_array_offset));
n_entries = le64toh(READ_NOW(d->data.n_entries));
/* First, seek by time */
r = find_data_object_by_boot_id(f, boot_id, &o, &b);
if (r < 0)
return r;
if (r == 0)
return -ENOENT;
r = generic_array_bisect_plus_one(f,
le64toh(o->data.entry_offset),
le64toh(o->data.entry_array_offset),
le64toh(o->data.n_entries),
monotonic,
test_object_monotonic,
direction,
NULL, &z, NULL);
if (r <= 0)
return r;
/* And now, continue seeking until we find an entry that
* exists in both bisection arrays */
r = journal_file_move_to_object(f, OBJECT_DATA, b, &o);
if (r < 0)
return r;
for (;;) {
uint64_t p, q;
r = generic_array_bisect_plus_one(f,
entry_offset,
entry_array_offset,
n_entries,
z,
test_object_offset,
direction,
NULL, &p, NULL);
if (r <= 0)
return r;
r = generic_array_bisect_plus_one(f,
le64toh(o->data.entry_offset),
le64toh(o->data.entry_array_offset),
le64toh(o->data.n_entries),
p,
test_object_offset,
direction,
NULL, &q, NULL);
if (r <= 0)
return r;
if (p == q) {
if (ret_object) {
r = journal_file_move_to_object(f, OBJECT_ENTRY, q, ret_object);
if (r < 0)
return r;
}
if (ret_offset)
*ret_offset = q;
return 1;
}
z = q;
}
}
int journal_file_move_to_entry_by_seqnum_for_data(
JournalFile *f,
Object *d,
uint64_t seqnum,
direction_t direction,
Object **ret_object,
uint64_t *ret_offset) {
assert(f);
assert(d);
assert(d->object.type == OBJECT_DATA);
return generic_array_bisect_plus_one(
f,
le64toh(d->data.entry_offset),
le64toh(d->data.entry_array_offset),
le64toh(d->data.n_entries),
seqnum,
test_object_seqnum,
direction,
ret_object, ret_offset, NULL);
}
int journal_file_move_to_entry_by_realtime_for_data(
JournalFile *f,
Object *d,
uint64_t realtime,
direction_t direction,
Object **ret, uint64_t *ret_offset) {
assert(f);
assert(d);
assert(d->object.type == OBJECT_DATA);
return generic_array_bisect_plus_one(
f,
le64toh(d->data.entry_offset),
le64toh(d->data.entry_array_offset),
le64toh(d->data.n_entries),
realtime,
test_object_realtime,
direction,
ret, ret_offset, NULL);
}
void journal_file_dump(JournalFile *f) {
Object *o;
uint64_t p;
int r;
assert(f);
assert(f->header);
journal_file_print_header(f);
p = le64toh(READ_NOW(f->header->header_size));
while (p != 0) {
const char *s;
Compression c;
r = journal_file_move_to_object(f, OBJECT_UNUSED, p, &o);
if (r < 0)
goto fail;
s = journal_object_type_to_string(o->object.type);
switch (o->object.type) {
case OBJECT_ENTRY:
assert(s);
printf("Type: %s seqnum=%"PRIu64" monotonic=%"PRIu64" realtime=%"PRIu64"\n",
s,
le64toh(o->entry.seqnum),
le64toh(o->entry.monotonic),
le64toh(o->entry.realtime));
break;
case OBJECT_TAG:
assert(s);
printf("Type: %s seqnum=%"PRIu64" epoch=%"PRIu64"\n",
s,
le64toh(o->tag.seqnum),
le64toh(o->tag.epoch));
break;
default:
if (s)
printf("Type: %s \n", s);
else
printf("Type: unknown (%i)", o->object.type);
break;
}
c = COMPRESSION_FROM_OBJECT(o);
if (c > COMPRESSION_NONE)
printf("Flags: %s\n",
compression_to_string(c));
if (p == le64toh(f->header->tail_object_offset))
p = 0;
else
p += ALIGN64(le64toh(o->object.size));
}
return;
fail:
log_error("File corrupt");
}
/* Note: the lifetime of the compound literal is the immediately surrounding block. */
#define FORMAT_TIMESTAMP_SAFE(t) (FORMAT_TIMESTAMP(t) ?: " --- ")
void journal_file_print_header(JournalFile *f) {
struct stat st;
assert(f);
assert(f->header);
printf("File path: %s\n"
"File ID: %s\n"
"Machine ID: %s\n"
"Boot ID: %s\n"
"Sequential number ID: %s\n"
"State: %s\n"
"Compatible flags:%s%s\n"
"Incompatible flags:%s%s%s%s%s%s\n"
"Header size: %"PRIu64"\n"
"Arena size: %"PRIu64"\n"
"Data hash table size: %"PRIu64"\n"
"Field hash table size: %"PRIu64"\n"
"Rotate suggested: %s\n"
"Head sequential number: %"PRIu64" (%"PRIx64")\n"
"Tail sequential number: %"PRIu64" (%"PRIx64")\n"
"Head realtime timestamp: %s (%"PRIx64")\n"
"Tail realtime timestamp: %s (%"PRIx64")\n"
"Tail monotonic timestamp: %s (%"PRIx64")\n"
"Objects: %"PRIu64"\n"
"Entry objects: %"PRIu64"\n",
f->path,
SD_ID128_TO_STRING(f->header->file_id),
SD_ID128_TO_STRING(f->header->machine_id),
SD_ID128_TO_STRING(f->header->boot_id),
SD_ID128_TO_STRING(f->header->seqnum_id),
f->header->state == STATE_OFFLINE ? "OFFLINE" :
f->header->state == STATE_ONLINE ? "ONLINE" :
f->header->state == STATE_ARCHIVED ? "ARCHIVED" : "UNKNOWN",
JOURNAL_HEADER_SEALED(f->header) ? " SEALED" : "",
(le32toh(f->header->compatible_flags) & ~HEADER_COMPATIBLE_ANY) ? " ???" : "",
JOURNAL_HEADER_COMPRESSED_XZ(f->header) ? " COMPRESSED-XZ" : "",
JOURNAL_HEADER_COMPRESSED_LZ4(f->header) ? " COMPRESSED-LZ4" : "",
JOURNAL_HEADER_COMPRESSED_ZSTD(f->header) ? " COMPRESSED-ZSTD" : "",
JOURNAL_HEADER_KEYED_HASH(f->header) ? " KEYED-HASH" : "",
JOURNAL_HEADER_COMPACT(f->header) ? " COMPACT" : "",
(le32toh(f->header->incompatible_flags) & ~HEADER_INCOMPATIBLE_ANY) ? " ???" : "",
le64toh(f->header->header_size),
le64toh(f->header->arena_size),
le64toh(f->header->data_hash_table_size) / sizeof(HashItem),
le64toh(f->header->field_hash_table_size) / sizeof(HashItem),
yes_no(journal_file_rotate_suggested(f, 0, LOG_DEBUG)),
le64toh(f->header->head_entry_seqnum), le64toh(f->header->head_entry_seqnum),
le64toh(f->header->tail_entry_seqnum), le64toh(f->header->tail_entry_seqnum),
FORMAT_TIMESTAMP_SAFE(le64toh(f->header->head_entry_realtime)), le64toh(f->header->head_entry_realtime),
FORMAT_TIMESTAMP_SAFE(le64toh(f->header->tail_entry_realtime)), le64toh(f->header->tail_entry_realtime),
FORMAT_TIMESPAN(le64toh(f->header->tail_entry_monotonic), USEC_PER_MSEC), le64toh(f->header->tail_entry_monotonic),
le64toh(f->header->n_objects),
le64toh(f->header->n_entries));
if (JOURNAL_HEADER_CONTAINS(f->header, n_data))
printf("Data objects: %"PRIu64"\n"
"Data hash table fill: %.1f%%\n",
le64toh(f->header->n_data),
100.0 * (double) le64toh(f->header->n_data) / ((double) (le64toh(f->header->data_hash_table_size) / sizeof(HashItem))));
if (JOURNAL_HEADER_CONTAINS(f->header, n_fields))
printf("Field objects: %"PRIu64"\n"
"Field hash table fill: %.1f%%\n",
le64toh(f->header->n_fields),
100.0 * (double) le64toh(f->header->n_fields) / ((double) (le64toh(f->header->field_hash_table_size) / sizeof(HashItem))));
if (JOURNAL_HEADER_CONTAINS(f->header, n_tags))
printf("Tag objects: %"PRIu64"\n",
le64toh(f->header->n_tags));
if (JOURNAL_HEADER_CONTAINS(f->header, n_entry_arrays))
printf("Entry array objects: %"PRIu64"\n",
le64toh(f->header->n_entry_arrays));
if (JOURNAL_HEADER_CONTAINS(f->header, field_hash_chain_depth))
printf("Deepest field hash chain: %" PRIu64"\n",
f->header->field_hash_chain_depth);
if (JOURNAL_HEADER_CONTAINS(f->header, data_hash_chain_depth))
printf("Deepest data hash chain: %" PRIu64"\n",
f->header->data_hash_chain_depth);
if (fstat(f->fd, &st) >= 0)
printf("Disk usage: %s\n", FORMAT_BYTES((uint64_t) st.st_blocks * 512ULL));
}
static int journal_file_warn_btrfs(JournalFile *f) {
unsigned attrs;
int r;
assert(f);
/* Before we write anything, check if the COW logic is turned
* off on btrfs. Given our write pattern that is quite
* unfriendly to COW file systems this should greatly improve
* performance on COW file systems, such as btrfs, at the
* expense of data integrity features (which shouldn't be too
* bad, given that we do our own checksumming). */
r = fd_is_fs_type(f->fd, BTRFS_SUPER_MAGIC);
if (r < 0)
return log_ratelimit_warning_errno(r, JOURNAL_LOG_RATELIMIT, "Failed to determine if journal is on btrfs: %m");
if (!r)
return 0;
r = read_attr_fd(f->fd, &attrs);
if (r < 0)
return log_ratelimit_warning_errno(r, JOURNAL_LOG_RATELIMIT, "Failed to read file attributes: %m");
if (attrs & FS_NOCOW_FL) {
log_debug("Detected btrfs file system with copy-on-write disabled, all is good.");
return 0;
}
log_ratelimit_notice(JOURNAL_LOG_RATELIMIT,
"Creating journal file %s on a btrfs file system, and copy-on-write is enabled. "
"This is likely to slow down journal access substantially, please consider turning "
"off the copy-on-write file attribute on the journal directory, using chattr +C.",
f->path);
return 1;
}
static void journal_default_metrics(JournalMetrics *m, int fd, bool compact) {
struct statvfs ss;
uint64_t fs_size = 0;
assert(m);
assert(fd >= 0);
if (fstatvfs(fd, &ss) >= 0)
fs_size = ss.f_frsize * ss.f_blocks;
else
log_debug_errno(errno, "Failed to determine disk size: %m");
if (m->max_use == UINT64_MAX) {
if (fs_size > 0)
m->max_use = CLAMP(PAGE_ALIGN(fs_size / 10), /* 10% of file system size */
MAX_USE_LOWER, MAX_USE_UPPER);
else
m->max_use = MAX_USE_LOWER;
} else {
m->max_use = PAGE_ALIGN(m->max_use);
if (m->max_use != 0 && m->max_use < JOURNAL_FILE_SIZE_MIN*2)
m->max_use = JOURNAL_FILE_SIZE_MIN*2;
}
if (m->min_use == UINT64_MAX) {
if (fs_size > 0)
m->min_use = CLAMP(PAGE_ALIGN(fs_size / 50), /* 2% of file system size */
MIN_USE_LOW, MIN_USE_HIGH);
else
m->min_use = MIN_USE_LOW;
}
if (m->min_use > m->max_use)
m->min_use = m->max_use;
if (m->max_size == UINT64_MAX)
m->max_size = MIN(PAGE_ALIGN(m->max_use / 8), /* 8 chunks */
MAX_SIZE_UPPER);
else
m->max_size = PAGE_ALIGN(m->max_size);
if (compact && m->max_size > JOURNAL_COMPACT_SIZE_MAX)
m->max_size = JOURNAL_COMPACT_SIZE_MAX;
if (m->max_size != 0) {
if (m->max_size < JOURNAL_FILE_SIZE_MIN)
m->max_size = JOURNAL_FILE_SIZE_MIN;
if (m->max_use != 0 && m->max_size*2 > m->max_use)
m->max_use = m->max_size*2;
}
if (m->min_size == UINT64_MAX)
m->min_size = JOURNAL_FILE_SIZE_MIN;
else
m->min_size = CLAMP(PAGE_ALIGN(m->min_size),
JOURNAL_FILE_SIZE_MIN,
m->max_size ?: UINT64_MAX);
if (m->keep_free == UINT64_MAX) {
if (fs_size > 0)
m->keep_free = MIN(PAGE_ALIGN(fs_size / 20), /* 5% of file system size */
KEEP_FREE_UPPER);
else
m->keep_free = DEFAULT_KEEP_FREE;
}
if (m->n_max_files == UINT64_MAX)
m->n_max_files = DEFAULT_N_MAX_FILES;
log_debug("Fixed min_use=%s max_use=%s max_size=%s min_size=%s keep_free=%s n_max_files=%" PRIu64,
FORMAT_BYTES(m->min_use),
FORMAT_BYTES(m->max_use),
FORMAT_BYTES(m->max_size),
FORMAT_BYTES(m->min_size),
FORMAT_BYTES(m->keep_free),
m->n_max_files);
}
int journal_file_open(
int fd,
const char *fname,
int open_flags,
JournalFileFlags file_flags,
mode_t mode,
uint64_t compress_threshold_bytes,
JournalMetrics *metrics,
MMapCache *mmap_cache,
JournalFile *template,
JournalFile **ret) {
bool newly_created = false;
JournalFile *f;
void *h;
int r;
assert(fd >= 0 || fname);
assert(file_flags >= 0);
assert(file_flags <= _JOURNAL_FILE_FLAGS_MAX);
assert(mmap_cache);
assert(ret);
if (!IN_SET((open_flags & O_ACCMODE), O_RDONLY, O_RDWR))
return -EINVAL;
if ((open_flags & O_ACCMODE) == O_RDONLY && FLAGS_SET(open_flags, O_CREAT))
return -EINVAL;
if (fname && (open_flags & O_CREAT) && !endswith(fname, ".journal"))
return -EINVAL;
f = new(JournalFile, 1);
if (!f)
return -ENOMEM;
*f = (JournalFile) {
.fd = fd,
.mode = mode,
.open_flags = open_flags,
.compress_threshold_bytes = compress_threshold_bytes == UINT64_MAX ?
DEFAULT_COMPRESS_THRESHOLD :
MAX(MIN_COMPRESS_THRESHOLD, compress_threshold_bytes),
.strict_order = FLAGS_SET(file_flags, JOURNAL_STRICT_ORDER),
};
if (fname) {
f->path = strdup(fname);
if (!f->path) {
r = -ENOMEM;
goto fail;
}
} else {
assert(fd >= 0);
/* If we don't know the path, fill in something explanatory and vaguely useful */
if (asprintf(&f->path, "/proc/self/%i", fd) < 0) {
r = -ENOMEM;
goto fail;
}
}
f->chain_cache = ordered_hashmap_new(&uint64_hash_ops);
if (!f->chain_cache) {
r = -ENOMEM;
goto fail;
}
if (f->fd < 0) {
/* We pass O_NONBLOCK here, so that in case somebody pointed us to some character device node or FIFO
* or so, we likely fail quickly than block for long. For regular files O_NONBLOCK has no effect, hence
* it doesn't hurt in that case. */
f->fd = openat_report_new(AT_FDCWD, f->path, f->open_flags|O_CLOEXEC|O_NONBLOCK, f->mode, &newly_created);
if (f->fd < 0) {
r = f->fd;
goto fail;
}
/* fds we opened here by us should also be closed by us. */
f->close_fd = true;
r = fd_nonblock(f->fd, false);
if (r < 0)
goto fail;
if (!newly_created) {
r = journal_file_fstat(f);
if (r < 0)
goto fail;
}
} else {
r = journal_file_fstat(f);
if (r < 0)
goto fail;
/* If we just got the fd passed in, we don't really know if we created the file anew */
newly_created = f->last_stat.st_size == 0 && journal_file_writable(f);
}
f->cache_fd = mmap_cache_add_fd(mmap_cache, f->fd, mmap_prot_from_open_flags(open_flags));
if (!f->cache_fd) {
r = -ENOMEM;
goto fail;
}
if (newly_created) {
(void) journal_file_warn_btrfs(f);
/* Let's attach the creation time to the journal file, so that the vacuuming code knows the age of this
* file even if the file might end up corrupted one day... Ideally we'd just use the creation time many
* file systems maintain for each file, but the API to query this is very new, hence let's emulate this
* via extended attributes. If extended attributes are not supported we'll just skip this, and rely
* solely on mtime/atime/ctime of the file. */
(void) fd_setcrtime(f->fd, 0);
r = journal_file_init_header(f, file_flags, template);
if (r < 0)
goto fail;
r = journal_file_fstat(f);
if (r < 0)
goto fail;
}
if (f->last_stat.st_size < (off_t) HEADER_SIZE_MIN) {
r = -ENODATA;
goto fail;
}
r = mmap_cache_fd_get(f->cache_fd, CONTEXT_HEADER, true, 0, PAGE_ALIGN(sizeof(Header)), &f->last_stat, &h);
if (r == -EINVAL) {
/* Some file systems (jffs2 or p9fs) don't support mmap() properly (or only read-only
* mmap()), and return EINVAL in that case. Let's propagate that as a more recognizable error
* code. */
r = -EAFNOSUPPORT;
goto fail;
}
if (r < 0)
goto fail;
f->header = h;
if (!newly_created) {
r = journal_file_verify_header(f);
if (r < 0)
goto fail;
}
#if HAVE_GCRYPT
if (!newly_created && journal_file_writable(f) && JOURNAL_HEADER_SEALED(f->header)) {
r = journal_file_fss_load(f);
if (r < 0)
goto fail;
}
#endif
if (journal_file_writable(f)) {
if (metrics) {
journal_default_metrics(metrics, f->fd, JOURNAL_HEADER_COMPACT(f->header));
f->metrics = *metrics;
} else if (template)
f->metrics = template->metrics;
r = journal_file_refresh_header(f);
if (r < 0)
goto fail;
}
#if HAVE_GCRYPT
r = journal_file_hmac_setup(f);
if (r < 0)
goto fail;
#endif
if (newly_created) {
r = journal_file_setup_field_hash_table(f);
if (r < 0)
goto fail;
r = journal_file_setup_data_hash_table(f);
if (r < 0)
goto fail;
#if HAVE_GCRYPT
r = journal_file_append_first_tag(f);
if (r < 0)
goto fail;
#endif
}
if (mmap_cache_fd_got_sigbus(f->cache_fd)) {
r = -EIO;
goto fail;
}
if (template && template->post_change_timer) {
r = journal_file_enable_post_change_timer(
f,
sd_event_source_get_event(template->post_change_timer),
template->post_change_timer_period);
if (r < 0)
goto fail;
}
/* The file is opened now successfully, thus we take possession of any passed in fd. */
f->close_fd = true;
if (DEBUG_LOGGING) {
static int last_seal = -1, last_compress = -1, last_keyed_hash = -1;
static uint64_t last_bytes = UINT64_MAX;
if (last_seal != JOURNAL_HEADER_SEALED(f->header) ||
last_keyed_hash != JOURNAL_HEADER_KEYED_HASH(f->header) ||
last_compress != JOURNAL_FILE_COMPRESS(f) ||
last_bytes != f->compress_threshold_bytes) {
log_debug("Journal effective settings seal=%s keyed_hash=%s compress=%s compress_threshold_bytes=%s",
yes_no(JOURNAL_HEADER_SEALED(f->header)), yes_no(JOURNAL_HEADER_KEYED_HASH(f->header)),
yes_no(JOURNAL_FILE_COMPRESS(f)), FORMAT_BYTES(f->compress_threshold_bytes));
last_seal = JOURNAL_HEADER_SEALED(f->header);
last_keyed_hash = JOURNAL_HEADER_KEYED_HASH(f->header);
last_compress = JOURNAL_FILE_COMPRESS(f);
last_bytes = f->compress_threshold_bytes;
}
}
*ret = f;
return 0;
fail:
if (f->cache_fd && mmap_cache_fd_got_sigbus(f->cache_fd))
r = -EIO;
(void) journal_file_close(f);
if (newly_created && fd < 0)
(void) unlink(fname);
return r;
}
int journal_file_parse_uid_from_filename(const char *path, uid_t *ret_uid) {
_cleanup_free_ char *buf = NULL, *p = NULL;
const char *a, *b, *at;
int r;
/* This helper returns -EREMOTE when the filename doesn't match user online/offline journal
* pattern. Hence it currently doesn't parse archived or disposed user journals. */
assert(path);
assert(ret_uid);
r = path_extract_filename(path, &p);
if (r < 0)
return r;
if (r == O_DIRECTORY)
return -EISDIR;
a = startswith(p, "user-");
if (!a)
return -EREMOTE;
b = endswith(p, ".journal");
if (!b)
return -EREMOTE;
at = strchr(a, '@');
if (at)
return -EREMOTE;
buf = strndup(a, b-a);
if (!buf)
return -ENOMEM;
return parse_uid(buf, ret_uid);
}
int journal_file_archive(JournalFile *f, char **ret_previous_path) {
_cleanup_free_ char *p = NULL;
assert(f);
if (!journal_file_writable(f))
return -EINVAL;
/* Is this a journal file that was passed to us as fd? If so, we synthesized a path name for it, and we refuse
* rotation, since we don't know the actual path, and couldn't rename the file hence. */
if (path_startswith(f->path, "/proc/self/fd"))
return -EINVAL;
if (!endswith(f->path, ".journal"))
return -EINVAL;
if (asprintf(&p, "%.*s@" SD_ID128_FORMAT_STR "-%016"PRIx64"-%016"PRIx64".journal",
(int) strlen(f->path) - 8, f->path,
SD_ID128_FORMAT_VAL(f->header->seqnum_id),
le64toh(f->header->head_entry_seqnum),
le64toh(f->header->head_entry_realtime)) < 0)
return -ENOMEM;
/* Try to rename the file to the archived version. If the file already was deleted, we'll get ENOENT, let's
* ignore that case. */
if (rename(f->path, p) < 0 && errno != ENOENT)
return -errno;
/* Sync the rename to disk */
(void) fsync_directory_of_file(f->fd);
if (ret_previous_path)
*ret_previous_path = f->path;
else
free(f->path);
f->path = TAKE_PTR(p);
/* Set as archive so offlining commits w/state=STATE_ARCHIVED. Previously we would set old_file->header->state
* to STATE_ARCHIVED directly here, but journal_file_set_offline() short-circuits when state != STATE_ONLINE,
* which would result in the rotated journal never getting fsync() called before closing. Now we simply queue
* the archive state by setting an archive bit, leaving the state as STATE_ONLINE so proper offlining
* occurs. */
f->archive = true;
return 0;
}
int journal_file_dispose(int dir_fd, const char *fname) {
_cleanup_free_ char *p = NULL;
assert(fname);
/* Renames a journal file to *.journal~, i.e. to mark it as corrupted or otherwise uncleanly shutdown. Note that
* this is done without looking into the file or changing any of its contents. The idea is that this is called
* whenever something is suspicious and we want to move the file away and make clear that it is not accessed
* for writing anymore. */
if (!endswith(fname, ".journal"))
return -EINVAL;
if (asprintf(&p, "%.*s@%016" PRIx64 "-%016" PRIx64 ".journal~",
(int) strlen(fname) - 8, fname,
now(CLOCK_REALTIME),
random_u64()) < 0)
return -ENOMEM;
if (renameat(dir_fd, fname, dir_fd, p) < 0)
return -errno;
return 0;
}
int journal_file_copy_entry(JournalFile *from, JournalFile *to, Object *o, uint64_t p) {
_cleanup_free_ EntryItem *items_alloc = NULL;
EntryItem *items;
uint64_t q, n, xor_hash = 0;
const sd_id128_t *boot_id;
dual_timestamp ts;
int r;
assert(from);
assert(to);
assert(o);
assert(p > 0);
if (!journal_file_writable(to))
return -EPERM;
ts = (dual_timestamp) {
.monotonic = le64toh(o->entry.monotonic),
.realtime = le64toh(o->entry.realtime),
};
boot_id = &o->entry.boot_id;
n = journal_file_entry_n_items(from, o);
if (n < ALLOCA_MAX / sizeof(EntryItem) / 2)
items = newa(EntryItem, n);
else {
items_alloc = new(EntryItem, n);
if (!items_alloc)
return -ENOMEM;
items = items_alloc;
}
for (uint64_t i = 0; i < n; i++) {
uint64_t h;
void *data;
size_t l;
Object *u;
q = journal_file_entry_item_object_offset(from, o, i);
r = journal_file_data_payload(from, NULL, q, NULL, 0, 0, &data, &l);
if (IN_SET(r, -EADDRNOTAVAIL, -EBADMSG)) {
log_debug_errno(r, "Entry item %"PRIu64" data object is bad, skipping over it: %m", i);
goto next;
}
if (r < 0)
return r;
assert(r > 0);
if (l == 0)
return -EBADMSG;
r = journal_file_append_data(to, data, l, &u, &h);
if (r < 0)
return r;
if (JOURNAL_HEADER_KEYED_HASH(to->header))
xor_hash ^= jenkins_hash64(data, l);
else
xor_hash ^= le64toh(u->data.hash);
items[i] = (EntryItem) {
.object_offset = h,
.hash = le64toh(u->data.hash),
};
next:
/* The above journal_file_data_payload() may clear or overwrite cached object. Hence, we need
* to re-read the object from the cache. */
r = journal_file_move_to_object(from, OBJECT_ENTRY, p, &o);
if (r < 0)
return r;
}
r = journal_file_append_entry_internal(to, &ts, boot_id, xor_hash, items, n, NULL, NULL, NULL);
if (mmap_cache_fd_got_sigbus(to->cache_fd))
return -EIO;
return r;
}
void journal_reset_metrics(JournalMetrics *m) {
assert(m);
/* Set everything to "pick automatic values". */
*m = (JournalMetrics) {
.min_use = UINT64_MAX,
.max_use = UINT64_MAX,
.min_size = UINT64_MAX,
.max_size = UINT64_MAX,
.keep_free = UINT64_MAX,
.n_max_files = UINT64_MAX,
};
}
int journal_file_get_cutoff_realtime_usec(JournalFile *f, usec_t *ret_from, usec_t *ret_to) {
assert(f);
assert(f->header);
assert(ret_from || ret_to);
if (ret_from) {
if (f->header->head_entry_realtime == 0)
return -ENOENT;
*ret_from = le64toh(f->header->head_entry_realtime);
}
if (ret_to) {
if (f->header->tail_entry_realtime == 0)
return -ENOENT;
*ret_to = le64toh(f->header->tail_entry_realtime);
}
return 1;
}
int journal_file_get_cutoff_monotonic_usec(JournalFile *f, sd_id128_t boot_id, usec_t *ret_from, usec_t *ret_to) {
Object *o;
uint64_t p;
int r;
assert(f);
assert(ret_from || ret_to);
/* FIXME: fix return value assignment on success with 0. */
r = find_data_object_by_boot_id(f, boot_id, &o, &p);
if (r <= 0)
return r;
if (le64toh(o->data.n_entries) <= 0)
return 0;
if (ret_from) {
r = journal_file_move_to_object(f, OBJECT_ENTRY, le64toh(o->data.entry_offset), &o);
if (r < 0)
return r;
*ret_from = le64toh(o->entry.monotonic);
}
if (ret_to) {
r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
if (r < 0)
return r;
r = generic_array_get_plus_one(f,
le64toh(o->data.entry_offset),
le64toh(o->data.entry_array_offset),
le64toh(o->data.n_entries) - 1,
DIRECTION_UP,
&o, NULL);
if (r <= 0)
return r;
*ret_to = le64toh(o->entry.monotonic);
}
return 1;
}
bool journal_file_rotate_suggested(JournalFile *f, usec_t max_file_usec, int log_level) {
assert(f);
assert(f->header);
/* If we gained new header fields we gained new features,
* hence suggest a rotation */
if (le64toh(f->header->header_size) < sizeof(Header)) {
log_ratelimit_full(log_level, JOURNAL_LOG_RATELIMIT,
"%s uses an outdated header, suggesting rotation.", f->path);
return true;
}
/* Let's check if the hash tables grew over a certain fill level (75%, borrowing this value from
* Java's hash table implementation), and if so suggest a rotation. To calculate the fill level we
* need the n_data field, which only exists in newer versions. */
if (JOURNAL_HEADER_CONTAINS(f->header, n_data))
if (le64toh(f->header->n_data) * 4ULL > (le64toh(f->header->data_hash_table_size) / sizeof(HashItem)) * 3ULL) {
log_ratelimit_full(
log_level, JOURNAL_LOG_RATELIMIT,
"Data hash table of %s has a fill level at %.1f (%"PRIu64" of %"PRIu64" items, %"PRIu64" file size, %"PRIu64" bytes per hash table item), suggesting rotation.",
f->path,
100.0 * (double) le64toh(f->header->n_data) / ((double) (le64toh(f->header->data_hash_table_size) / sizeof(HashItem))),
le64toh(f->header->n_data),
le64toh(f->header->data_hash_table_size) / sizeof(HashItem),
(uint64_t) f->last_stat.st_size,
f->last_stat.st_size / le64toh(f->header->n_data));
return true;
}
if (JOURNAL_HEADER_CONTAINS(f->header, n_fields))
if (le64toh(f->header->n_fields) * 4ULL > (le64toh(f->header->field_hash_table_size) / sizeof(HashItem)) * 3ULL) {
log_ratelimit_full(
log_level, JOURNAL_LOG_RATELIMIT,
"Field hash table of %s has a fill level at %.1f (%"PRIu64" of %"PRIu64" items), suggesting rotation.",
f->path,
100.0 * (double) le64toh(f->header->n_fields) / ((double) (le64toh(f->header->field_hash_table_size) / sizeof(HashItem))),
le64toh(f->header->n_fields),
le64toh(f->header->field_hash_table_size) / sizeof(HashItem));
return true;
}
/* If there are too many hash collisions somebody is most likely playing games with us. Hence, if our
* longest chain is longer than some threshold, let's suggest rotation. */
if (JOURNAL_HEADER_CONTAINS(f->header, data_hash_chain_depth) &&
le64toh(f->header->data_hash_chain_depth) > HASH_CHAIN_DEPTH_MAX) {
log_ratelimit_full(
log_level, JOURNAL_LOG_RATELIMIT,
"Data hash table of %s has deepest hash chain of length %" PRIu64 ", suggesting rotation.",
f->path, le64toh(f->header->data_hash_chain_depth));
return true;
}
if (JOURNAL_HEADER_CONTAINS(f->header, field_hash_chain_depth) &&
le64toh(f->header->field_hash_chain_depth) > HASH_CHAIN_DEPTH_MAX) {
log_ratelimit_full(
log_level, JOURNAL_LOG_RATELIMIT,
"Field hash table of %s has deepest hash chain of length at %" PRIu64 ", suggesting rotation.",
f->path, le64toh(f->header->field_hash_chain_depth));
return true;
}
/* Are the data objects properly indexed by field objects? */
if (JOURNAL_HEADER_CONTAINS(f->header, n_data) &&
JOURNAL_HEADER_CONTAINS(f->header, n_fields) &&
le64toh(f->header->n_data) > 0 &&
le64toh(f->header->n_fields) == 0) {
log_ratelimit_full(
log_level, JOURNAL_LOG_RATELIMIT,
"Data objects of %s are not indexed by field objects, suggesting rotation.",
f->path);
return true;
}
if (max_file_usec > 0) {
usec_t t, h;
h = le64toh(f->header->head_entry_realtime);
t = now(CLOCK_REALTIME);
if (h > 0 && t > h + max_file_usec) {
log_ratelimit_full(
log_level, JOURNAL_LOG_RATELIMIT,
"Oldest entry in %s is older than the configured file retention duration (%s), suggesting rotation.",
f->path, FORMAT_TIMESPAN(max_file_usec, USEC_PER_SEC));
return true;
}
}
return false;
}
static const char * const journal_object_type_table[] = {
[OBJECT_UNUSED] = "unused",
[OBJECT_DATA] = "data",
[OBJECT_FIELD] = "field",
[OBJECT_ENTRY] = "entry",
[OBJECT_DATA_HASH_TABLE] = "data hash table",
[OBJECT_FIELD_HASH_TABLE] = "field hash table",
[OBJECT_ENTRY_ARRAY] = "entry array",
[OBJECT_TAG] = "tag",
};
DEFINE_STRING_TABLE_LOOKUP_TO_STRING(journal_object_type, ObjectType);