2011-10-08 02:20:44 +02:00
|
|
|
/*-*- Mode: C; c-basic-offset: 8; indent-tabs-mode: nil -*-*/
|
|
|
|
|
|
|
|
/***
|
|
|
|
This file is part of systemd.
|
|
|
|
|
|
|
|
Copyright 2011 Lennart Poettering
|
|
|
|
|
|
|
|
systemd is free software; you can redistribute it and/or modify it
|
2012-04-12 00:20:58 +02:00
|
|
|
under the terms of the GNU Lesser General Public License as published by
|
|
|
|
the Free Software Foundation; either version 2.1 of the License, or
|
2011-10-08 02:20:44 +02:00
|
|
|
(at your option) any later version.
|
|
|
|
|
|
|
|
systemd is distributed in the hope that it will be useful, but
|
|
|
|
WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
|
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
2012-04-12 00:20:58 +02:00
|
|
|
Lesser General Public License for more details.
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2012-04-12 00:20:58 +02:00
|
|
|
You should have received a copy of the GNU Lesser General Public License
|
2011-10-08 02:20:44 +02:00
|
|
|
along with systemd; If not, see <http://www.gnu.org/licenses/>.
|
|
|
|
***/
|
|
|
|
|
|
|
|
#include <sys/mman.h>
|
|
|
|
#include <errno.h>
|
|
|
|
#include <sys/uio.h>
|
|
|
|
#include <unistd.h>
|
|
|
|
#include <sys/statvfs.h>
|
|
|
|
#include <fcntl.h>
|
|
|
|
#include <stddef.h>
|
|
|
|
|
|
|
|
#include "journal-def.h"
|
|
|
|
#include "journal-file.h"
|
|
|
|
#include "lookup3.h"
|
2011-12-21 02:40:59 +01:00
|
|
|
#include "compress.h"
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2012-07-17 00:36:15 +02:00
|
|
|
#define DEFAULT_DATA_HASH_TABLE_SIZE (2047ULL*sizeof(HashItem))
|
|
|
|
#define DEFAULT_FIELD_HASH_TABLE_SIZE (333ULL*sizeof(HashItem))
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2012-04-02 19:04:56 +02:00
|
|
|
#define DEFAULT_WINDOW_SIZE (8ULL*1024ULL*1024ULL)
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2012-01-31 20:35:07 +01:00
|
|
|
#define COMPRESSION_SIZE_THRESHOLD (512ULL)
|
2011-12-21 02:40:59 +01:00
|
|
|
|
2011-12-31 02:31:54 +01:00
|
|
|
/* This is the minimum journal file size */
|
2012-03-15 23:25:14 +01:00
|
|
|
#define JOURNAL_FILE_SIZE_MIN (64ULL*1024ULL) /* 64 KiB */
|
2011-12-31 02:31:54 +01:00
|
|
|
|
|
|
|
/* These are the lower and upper bounds if we deduce the max_use value
|
|
|
|
* from the file system size */
|
|
|
|
#define DEFAULT_MAX_USE_LOWER (1ULL*1024ULL*1024ULL) /* 1 MiB */
|
|
|
|
#define DEFAULT_MAX_USE_UPPER (4ULL*1024ULL*1024ULL*1024ULL) /* 4 GiB */
|
|
|
|
|
|
|
|
/* This is the upper bound if we deduce max_size from max_use */
|
2012-03-15 02:57:39 +01:00
|
|
|
#define DEFAULT_MAX_SIZE_UPPER (128ULL*1024ULL*1024ULL) /* 128 MiB */
|
2011-12-31 02:31:54 +01:00
|
|
|
|
|
|
|
/* This is the upper bound if we deduce the keep_free value from the
|
|
|
|
* file system size */
|
|
|
|
#define DEFAULT_KEEP_FREE_UPPER (4ULL*1024ULL*1024ULL*1024ULL) /* 4 GiB */
|
|
|
|
|
|
|
|
/* This is the keep_free value when we can't determine the system
|
|
|
|
* size */
|
|
|
|
#define DEFAULT_KEEP_FREE (1024ULL*1024ULL) /* 1 MB */
|
|
|
|
|
2012-07-16 22:24:02 +02:00
|
|
|
/* n_data was the first entry we added after the initial file format design */
|
|
|
|
#define HEADER_SIZE_MIN ALIGN64(offsetof(Header, n_data))
|
2011-10-08 02:20:44 +02:00
|
|
|
|
|
|
|
#define ALIGN64(x) (((x) + 7ULL) & ~7ULL)
|
|
|
|
|
2012-07-16 22:24:02 +02:00
|
|
|
#define JOURNAL_HEADER_CONTAINS(h, field) \
|
|
|
|
(le64toh((h)->header_size) >= offsetof(Header, field) + sizeof((h)->field))
|
|
|
|
|
|
|
|
static const char signature[] = { 'L', 'P', 'K', 'S', 'H', 'H', 'R', 'H' };
|
|
|
|
|
2011-10-08 02:20:44 +02:00
|
|
|
void journal_file_close(JournalFile *f) {
|
2011-11-08 18:20:03 +01:00
|
|
|
int t;
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
assert(f);
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2012-06-12 16:45:09 +02:00
|
|
|
if (f->header) {
|
2012-07-17 00:55:02 +02:00
|
|
|
/* Mark the file offline. Don't override the archived state if it already is set */
|
|
|
|
if (f->writable && f->header->state == STATE_ONLINE)
|
2012-06-12 16:45:09 +02:00
|
|
|
f->header->state = STATE_OFFLINE;
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2012-06-12 16:45:09 +02:00
|
|
|
munmap(f->header, PAGE_ALIGN(sizeof(Header)));
|
|
|
|
}
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
for (t = 0; t < _WINDOW_MAX; t++)
|
|
|
|
if (f->windows[t].ptr)
|
|
|
|
munmap(f->windows[t].ptr, f->windows[t].size);
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-10-13 05:19:35 +02:00
|
|
|
if (f->fd >= 0)
|
|
|
|
close_nointr_nofail(f->fd);
|
|
|
|
|
2011-10-08 02:20:44 +02:00
|
|
|
free(f->path);
|
2011-12-21 02:40:59 +01:00
|
|
|
|
|
|
|
#ifdef HAVE_XZ
|
|
|
|
free(f->compress_buffer);
|
|
|
|
#endif
|
|
|
|
|
2011-10-08 02:20:44 +02:00
|
|
|
free(f);
|
|
|
|
}
|
|
|
|
|
2011-10-13 05:19:35 +02:00
|
|
|
static int journal_file_init_header(JournalFile *f, JournalFile *template) {
|
2011-10-08 02:20:44 +02:00
|
|
|
Header h;
|
|
|
|
ssize_t k;
|
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(f);
|
|
|
|
|
|
|
|
zero(h);
|
|
|
|
memcpy(h.signature, signature, 8);
|
2012-06-05 20:32:55 +02:00
|
|
|
h.header_size = htole64(ALIGN64(sizeof(h)));
|
2011-10-08 02:20:44 +02:00
|
|
|
|
|
|
|
r = sd_id128_randomize(&h.file_id);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
2011-10-13 05:19:35 +02:00
|
|
|
if (template) {
|
|
|
|
h.seqnum_id = template->header->seqnum_id;
|
2012-07-16 22:24:02 +02:00
|
|
|
h.tail_seqnum = template->header->tail_seqnum;
|
2011-10-13 05:19:35 +02:00
|
|
|
} else
|
|
|
|
h.seqnum_id = h.file_id;
|
2011-10-08 02:20:44 +02:00
|
|
|
|
|
|
|
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;
|
2011-11-08 18:20:03 +01:00
|
|
|
sd_id128_t boot_id;
|
2011-10-08 02:20:44 +02:00
|
|
|
|
|
|
|
assert(f);
|
|
|
|
|
|
|
|
r = sd_id128_get_machine(&f->header->machine_id);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
r = sd_id128_get_boot(&boot_id);
|
2011-10-08 02:20:44 +02:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
if (sd_id128_equal(boot_id, f->header->boot_id))
|
|
|
|
f->tail_entry_monotonic_valid = true;
|
|
|
|
|
|
|
|
f->header->boot_id = boot_id;
|
|
|
|
|
|
|
|
f->header->state = STATE_ONLINE;
|
2012-01-04 02:15:45 +01:00
|
|
|
|
|
|
|
__sync_synchronize();
|
|
|
|
|
2011-10-08 02:20:44 +02:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int journal_file_verify_header(JournalFile *f) {
|
|
|
|
assert(f);
|
|
|
|
|
|
|
|
if (memcmp(f->header, signature, 8))
|
|
|
|
return -EBADMSG;
|
|
|
|
|
2011-12-21 02:40:59 +01:00
|
|
|
#ifdef HAVE_XZ
|
|
|
|
if ((le64toh(f->header->incompatible_flags) & ~HEADER_INCOMPATIBLE_COMPRESSED) != 0)
|
|
|
|
return -EPROTONOSUPPORT;
|
|
|
|
#else
|
2011-10-08 02:20:44 +02:00
|
|
|
if (f->header->incompatible_flags != 0)
|
|
|
|
return -EPROTONOSUPPORT;
|
2011-12-21 02:40:59 +01:00
|
|
|
#endif
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2012-07-16 22:24:02 +02:00
|
|
|
/* The first addition was n_data, so check that we are at least this large */
|
|
|
|
if (le64toh(f->header->header_size) < HEADER_SIZE_MIN)
|
2012-06-05 20:32:55 +02:00
|
|
|
return -EBADMSG;
|
|
|
|
|
|
|
|
if ((uint64_t) f->last_stat.st_size < (le64toh(f->header->header_size) + le64toh(f->header->arena_size)))
|
2011-10-08 02:20:44 +02:00
|
|
|
return -ENODATA;
|
|
|
|
|
|
|
|
if (f->writable) {
|
2012-03-15 17:50:03 +01:00
|
|
|
uint8_t state;
|
2011-10-08 02:20:44 +02:00
|
|
|
sd_id128_t machine_id;
|
|
|
|
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 -EHOSTDOWN;
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
state = f->header->state;
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2012-07-16 22:51:28 +02:00
|
|
|
if (state == STATE_ONLINE) {
|
|
|
|
log_debug("Journal file %s is already online. Assuming unclean closing.", f->path);
|
|
|
|
return -EBUSY;
|
|
|
|
} else if (state == STATE_ARCHIVED)
|
2011-10-08 02:20:44 +02:00
|
|
|
return -ESHUTDOWN;
|
2012-07-16 22:51:28 +02:00
|
|
|
else if (state != STATE_OFFLINE) {
|
|
|
|
log_debug("Journal file %s has unknown state %u.", f->path, state);
|
|
|
|
return -EBUSY;
|
|
|
|
}
|
2011-10-08 02:20:44 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int journal_file_allocate(JournalFile *f, uint64_t offset, uint64_t size) {
|
|
|
|
uint64_t old_size, new_size;
|
2012-05-09 20:43:34 +02:00
|
|
|
int r;
|
2011-10-08 02:20:44 +02:00
|
|
|
|
|
|
|
assert(f);
|
|
|
|
|
|
|
|
/* We assume that this file is not sparse, and we know that
|
2011-10-12 04:41:27 +02:00
|
|
|
* for sure, since we always call posix_fallocate()
|
2011-10-08 02:20:44 +02:00
|
|
|
* ourselves */
|
|
|
|
|
|
|
|
old_size =
|
2012-06-05 20:32:55 +02:00
|
|
|
le64toh(f->header->header_size) +
|
2011-10-08 02:20:44 +02:00
|
|
|
le64toh(f->header->arena_size);
|
|
|
|
|
2011-12-20 02:38:36 +01:00
|
|
|
new_size = PAGE_ALIGN(offset + size);
|
2012-06-05 20:32:55 +02:00
|
|
|
if (new_size < le64toh(f->header->header_size))
|
|
|
|
new_size = le64toh(f->header->header_size);
|
2011-12-20 02:38:36 +01:00
|
|
|
|
|
|
|
if (new_size <= old_size)
|
2011-10-08 02:20:44 +02:00
|
|
|
return 0;
|
|
|
|
|
2011-12-20 02:38:36 +01:00
|
|
|
if (f->metrics.max_size > 0 &&
|
|
|
|
new_size > f->metrics.max_size)
|
|
|
|
return -E2BIG;
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-12-20 02:38:36 +01:00
|
|
|
if (new_size > f->metrics.min_size &&
|
|
|
|
f->metrics.keep_free > 0) {
|
2011-10-08 02:20:44 +02:00
|
|
|
struct statvfs svfs;
|
|
|
|
|
|
|
|
if (fstatvfs(f->fd, &svfs) >= 0) {
|
|
|
|
uint64_t available;
|
|
|
|
|
|
|
|
available = svfs.f_bfree * svfs.f_bsize;
|
|
|
|
|
2011-12-20 02:38:36 +01:00
|
|
|
if (available >= f->metrics.keep_free)
|
|
|
|
available -= f->metrics.keep_free;
|
2011-10-08 02:20:44 +02:00
|
|
|
else
|
|
|
|
available = 0;
|
|
|
|
|
|
|
|
if (new_size - old_size > available)
|
|
|
|
return -E2BIG;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2011-12-20 02:38:36 +01:00
|
|
|
/* Note that the glibc fallocate() fallback is very
|
|
|
|
inefficient, hence we try to minimize the allocation area
|
|
|
|
as we can. */
|
2012-05-09 20:43:34 +02:00
|
|
|
r = posix_fallocate(f->fd, old_size, new_size - old_size);
|
|
|
|
if (r != 0)
|
|
|
|
return -r;
|
2011-10-08 02:20:44 +02:00
|
|
|
|
|
|
|
if (fstat(f->fd, &f->last_stat) < 0)
|
|
|
|
return -errno;
|
|
|
|
|
2012-06-05 20:32:55 +02:00
|
|
|
f->header->arena_size = htole64(new_size - le64toh(f->header->header_size));
|
2011-10-08 02:20:44 +02:00
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int journal_file_map(
|
|
|
|
JournalFile *f,
|
|
|
|
uint64_t offset,
|
|
|
|
uint64_t size,
|
|
|
|
void **_window,
|
|
|
|
uint64_t *_woffset,
|
|
|
|
uint64_t *_wsize,
|
|
|
|
void **ret) {
|
|
|
|
|
|
|
|
uint64_t woffset, wsize;
|
|
|
|
void *window;
|
|
|
|
|
|
|
|
assert(f);
|
|
|
|
assert(size > 0);
|
|
|
|
assert(ret);
|
|
|
|
|
|
|
|
woffset = offset & ~((uint64_t) page_size() - 1ULL);
|
|
|
|
wsize = size + (offset - woffset);
|
|
|
|
wsize = PAGE_ALIGN(wsize);
|
|
|
|
|
2011-12-28 01:53:06 +01:00
|
|
|
/* Avoid SIGBUS on invalid accesses */
|
|
|
|
if (woffset + wsize > (uint64_t) PAGE_ALIGN(f->last_stat.st_size))
|
|
|
|
return -EADDRNOTAVAIL;
|
|
|
|
|
2011-10-08 02:20:44 +02:00
|
|
|
window = mmap(NULL, wsize, f->prot, MAP_SHARED, f->fd, woffset);
|
|
|
|
if (window == MAP_FAILED)
|
|
|
|
return -errno;
|
|
|
|
|
|
|
|
if (_window)
|
|
|
|
*_window = window;
|
|
|
|
|
|
|
|
if (_woffset)
|
|
|
|
*_woffset = woffset;
|
|
|
|
|
|
|
|
if (_wsize)
|
|
|
|
*_wsize = wsize;
|
|
|
|
|
|
|
|
*ret = (uint8_t*) window + (offset - woffset);
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
static int journal_file_move_to(JournalFile *f, int wt, uint64_t offset, uint64_t size, void **ret) {
|
2011-12-31 03:55:35 +01:00
|
|
|
void *p = NULL;
|
2011-10-08 02:20:44 +02:00
|
|
|
uint64_t delta;
|
|
|
|
int r;
|
2011-11-08 18:20:03 +01:00
|
|
|
Window *w;
|
2011-10-08 02:20:44 +02:00
|
|
|
|
|
|
|
assert(f);
|
|
|
|
assert(ret);
|
2011-11-08 18:20:03 +01:00
|
|
|
assert(wt >= 0);
|
|
|
|
assert(wt < _WINDOW_MAX);
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2012-01-04 02:16:38 +01:00
|
|
|
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. */
|
|
|
|
|
|
|
|
if (fstat(f->fd, &f->last_stat) < 0 ||
|
|
|
|
offset + size > (uint64_t) f->last_stat.st_size)
|
|
|
|
return -EADDRNOTAVAIL;
|
|
|
|
}
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
w = f->windows + wt;
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
if (_likely_(w->ptr &&
|
|
|
|
w->offset <= offset &&
|
|
|
|
w->offset + w->size >= offset + size)) {
|
|
|
|
|
|
|
|
*ret = (uint8_t*) w->ptr + (offset - w->offset);
|
2011-10-08 02:20:44 +02:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
if (w->ptr) {
|
|
|
|
if (munmap(w->ptr, w->size) < 0)
|
2011-10-08 02:20:44 +02:00
|
|
|
return -errno;
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
w->ptr = NULL;
|
|
|
|
w->size = w->offset = 0;
|
2011-10-08 02:20:44 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
if (size < DEFAULT_WINDOW_SIZE) {
|
|
|
|
/* If the default window size is larger then what was
|
|
|
|
* asked for extend the mapping a bit in the hope to
|
|
|
|
* minimize needed remappings later on. We add half
|
|
|
|
* the window space before and half behind the
|
|
|
|
* requested mapping */
|
|
|
|
|
2012-01-11 03:55:24 +01:00
|
|
|
delta = (DEFAULT_WINDOW_SIZE - size) / 2;
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2012-01-04 02:21:04 +01:00
|
|
|
if (delta > offset)
|
2011-10-08 02:20:44 +02:00
|
|
|
delta = offset;
|
|
|
|
|
|
|
|
offset -= delta;
|
2012-01-04 02:21:04 +01:00
|
|
|
size = DEFAULT_WINDOW_SIZE;
|
2011-10-08 02:20:44 +02:00
|
|
|
} else
|
|
|
|
delta = 0;
|
|
|
|
|
2011-12-28 01:53:06 +01:00
|
|
|
if (offset + size > (uint64_t) f->last_stat.st_size)
|
2012-01-11 03:55:24 +01:00
|
|
|
size = (uint64_t) f->last_stat.st_size - offset;
|
2011-12-28 01:53:06 +01:00
|
|
|
|
|
|
|
if (size <= 0)
|
|
|
|
return -EADDRNOTAVAIL;
|
|
|
|
|
2011-10-08 02:20:44 +02:00
|
|
|
r = journal_file_map(f,
|
|
|
|
offset, size,
|
2011-11-08 18:20:03 +01:00
|
|
|
&w->ptr, &w->offset, &w->size,
|
|
|
|
&p);
|
2011-10-08 02:20:44 +02:00
|
|
|
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
|
|
|
*ret = (uint8_t*) p + delta;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static bool verify_hash(Object *o) {
|
2011-11-08 18:20:03 +01:00
|
|
|
uint64_t h1, h2;
|
2011-10-08 02:20:44 +02:00
|
|
|
|
|
|
|
assert(o);
|
|
|
|
|
2011-12-21 02:40:59 +01:00
|
|
|
if (o->object.type == OBJECT_DATA && !(o->object.flags & OBJECT_COMPRESSED)) {
|
2011-10-08 02:20:44 +02:00
|
|
|
h1 = le64toh(o->data.hash);
|
2011-11-08 18:20:03 +01:00
|
|
|
h2 = hash64(o->data.payload, le64toh(o->object.size) - offsetof(Object, data.payload));
|
|
|
|
} else if (o->object.type == OBJECT_FIELD) {
|
|
|
|
h1 = le64toh(o->field.hash);
|
|
|
|
h2 = hash64(o->field.payload, le64toh(o->object.size) - offsetof(Object, field.payload));
|
|
|
|
} else
|
|
|
|
return true;
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
return h1 == h2;
|
2011-10-08 02:20:44 +02:00
|
|
|
}
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
int journal_file_move_to_object(JournalFile *f, int type, uint64_t offset, Object **ret) {
|
2011-10-08 02:20:44 +02:00
|
|
|
int r;
|
|
|
|
void *t;
|
|
|
|
Object *o;
|
|
|
|
uint64_t s;
|
|
|
|
|
|
|
|
assert(f);
|
|
|
|
assert(ret);
|
2011-11-08 18:20:03 +01:00
|
|
|
assert(type < _OBJECT_TYPE_MAX);
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
r = journal_file_move_to(f, type >= 0 ? type : WINDOW_UNKNOWN, offset, sizeof(ObjectHeader), &t);
|
2011-10-08 02:20:44 +02:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
|
|
|
o = (Object*) t;
|
|
|
|
s = le64toh(o->object.size);
|
|
|
|
|
|
|
|
if (s < sizeof(ObjectHeader))
|
|
|
|
return -EBADMSG;
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
if (type >= 0 && o->object.type != type)
|
2011-10-08 02:20:44 +02:00
|
|
|
return -EBADMSG;
|
|
|
|
|
|
|
|
if (s > sizeof(ObjectHeader)) {
|
2011-11-08 18:20:03 +01:00
|
|
|
r = journal_file_move_to(f, o->object.type, offset, s, &t);
|
2011-10-08 02:20:44 +02:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
|
|
|
o = (Object*) t;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!verify_hash(o))
|
|
|
|
return -EBADMSG;
|
|
|
|
|
|
|
|
*ret = o;
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2011-10-14 05:12:58 +02:00
|
|
|
static uint64_t journal_file_seqnum(JournalFile *f, uint64_t *seqnum) {
|
2011-10-08 02:20:44 +02:00
|
|
|
uint64_t r;
|
|
|
|
|
|
|
|
assert(f);
|
|
|
|
|
2012-07-16 22:24:02 +02:00
|
|
|
r = le64toh(f->header->tail_seqnum) + 1;
|
2011-10-14 05:12:58 +02:00
|
|
|
|
|
|
|
if (seqnum) {
|
2011-11-08 18:20:03 +01:00
|
|
|
/* If an external seqnum counter was passed, we update
|
2011-10-14 05:12:58 +02:00
|
|
|
* both the local and the external one, and set it to
|
|
|
|
* the maximum of both */
|
|
|
|
|
|
|
|
if (*seqnum + 1 > r)
|
|
|
|
r = *seqnum + 1;
|
|
|
|
|
|
|
|
*seqnum = r;
|
|
|
|
}
|
|
|
|
|
2012-07-16 22:24:02 +02:00
|
|
|
f->header->tail_seqnum = htole64(r);
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2012-07-16 22:24:02 +02:00
|
|
|
if (f->header->head_seqnum == 0)
|
|
|
|
f->header->head_seqnum = htole64(r);
|
2011-11-08 18:20:03 +01:00
|
|
|
|
2011-10-08 02:20:44 +02:00
|
|
|
return r;
|
|
|
|
}
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
static int journal_file_append_object(JournalFile *f, int type, uint64_t size, Object **ret, uint64_t *offset) {
|
2011-10-08 02:20:44 +02:00
|
|
|
int r;
|
|
|
|
uint64_t p;
|
|
|
|
Object *tail, *o;
|
|
|
|
void *t;
|
|
|
|
|
|
|
|
assert(f);
|
|
|
|
assert(size >= sizeof(ObjectHeader));
|
|
|
|
assert(offset);
|
|
|
|
assert(ret);
|
|
|
|
|
|
|
|
p = le64toh(f->header->tail_object_offset);
|
|
|
|
if (p == 0)
|
2012-06-05 20:32:55 +02:00
|
|
|
p = le64toh(f->header->header_size);
|
2011-10-08 02:20:44 +02:00
|
|
|
else {
|
2011-11-08 18:20:03 +01:00
|
|
|
r = journal_file_move_to_object(f, -1, p, &tail);
|
2011-10-08 02:20:44 +02:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
|
|
|
p += ALIGN64(le64toh(tail->object.size));
|
|
|
|
}
|
|
|
|
|
|
|
|
r = journal_file_allocate(f, p, size);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
r = journal_file_move_to(f, type, p, size, &t);
|
2011-10-08 02:20:44 +02:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
|
|
|
o = (Object*) t;
|
|
|
|
|
|
|
|
zero(o->object);
|
2011-11-08 18:20:03 +01:00
|
|
|
o->object.type = type;
|
2011-10-08 02:20:44 +02:00
|
|
|
o->object.size = htole64(size);
|
|
|
|
|
|
|
|
f->header->tail_object_offset = htole64(p);
|
|
|
|
f->header->n_objects = htole64(le64toh(f->header->n_objects) + 1);
|
|
|
|
|
|
|
|
*ret = o;
|
|
|
|
*offset = p;
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
static int journal_file_setup_data_hash_table(JournalFile *f) {
|
2011-10-08 02:20:44 +02:00
|
|
|
uint64_t s, p;
|
|
|
|
Object *o;
|
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(f);
|
|
|
|
|
2012-07-18 16:22:25 +02:00
|
|
|
/* We estimate that we need 1 hash table entry per 768 of
|
2012-07-17 00:36:15 +02:00
|
|
|
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. */
|
|
|
|
|
2012-07-18 16:22:25 +02:00
|
|
|
s = (f->metrics.max_size * 4 / 768 / 3) * sizeof(HashItem);
|
2012-07-17 00:36:15 +02:00
|
|
|
if (s < DEFAULT_DATA_HASH_TABLE_SIZE)
|
|
|
|
s = DEFAULT_DATA_HASH_TABLE_SIZE;
|
|
|
|
|
2012-07-18 16:22:25 +02:00
|
|
|
log_info("Reserving %llu entries in hash table.", (unsigned long long) (s / sizeof(HashItem)));
|
2012-07-17 00:36:15 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
r = journal_file_append_object(f,
|
|
|
|
OBJECT_DATA_HASH_TABLE,
|
|
|
|
offsetof(Object, hash_table.items) + s,
|
|
|
|
&o, &p);
|
2011-10-08 02:20:44 +02:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
memset(o->hash_table.items, 0, s);
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
f->header->data_hash_table_offset = htole64(p + offsetof(Object, hash_table.items));
|
|
|
|
f->header->data_hash_table_size = htole64(s);
|
2011-10-08 02:20:44 +02:00
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
static int journal_file_setup_field_hash_table(JournalFile *f) {
|
2011-10-08 02:20:44 +02:00
|
|
|
uint64_t s, p;
|
|
|
|
Object *o;
|
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(f);
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
s = DEFAULT_FIELD_HASH_TABLE_SIZE;
|
|
|
|
r = journal_file_append_object(f,
|
|
|
|
OBJECT_FIELD_HASH_TABLE,
|
|
|
|
offsetof(Object, hash_table.items) + s,
|
|
|
|
&o, &p);
|
2011-10-08 02:20:44 +02:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
memset(o->hash_table.items, 0, s);
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
f->header->field_hash_table_offset = htole64(p + offsetof(Object, hash_table.items));
|
|
|
|
f->header->field_hash_table_size = htole64(s);
|
2011-10-08 02:20:44 +02:00
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
static int journal_file_map_data_hash_table(JournalFile *f) {
|
2011-10-08 02:20:44 +02:00
|
|
|
uint64_t s, p;
|
|
|
|
void *t;
|
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(f);
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
p = le64toh(f->header->data_hash_table_offset);
|
|
|
|
s = le64toh(f->header->data_hash_table_size);
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
r = journal_file_move_to(f,
|
|
|
|
WINDOW_DATA_HASH_TABLE,
|
|
|
|
p, s,
|
|
|
|
&t);
|
2011-10-08 02:20:44 +02:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
f->data_hash_table = t;
|
2011-10-08 02:20:44 +02:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
static int journal_file_map_field_hash_table(JournalFile *f) {
|
2011-10-08 02:20:44 +02:00
|
|
|
uint64_t s, p;
|
|
|
|
void *t;
|
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(f);
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
p = le64toh(f->header->field_hash_table_offset);
|
|
|
|
s = le64toh(f->header->field_hash_table_size);
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
r = journal_file_move_to(f,
|
|
|
|
WINDOW_FIELD_HASH_TABLE,
|
|
|
|
p, s,
|
|
|
|
&t);
|
2011-10-08 02:20:44 +02:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
f->field_hash_table = t;
|
2011-10-08 02:20:44 +02:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
static int journal_file_link_data(JournalFile *f, Object *o, uint64_t offset, uint64_t hash) {
|
|
|
|
uint64_t p, h;
|
2011-10-08 02:20:44 +02:00
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(f);
|
|
|
|
assert(o);
|
|
|
|
assert(offset > 0);
|
2011-11-08 18:20:03 +01:00
|
|
|
assert(o->object.type == OBJECT_DATA);
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2012-03-06 02:42:32 +01:00
|
|
|
/* This might alter the window we are looking at */
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
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;
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
h = hash % (le64toh(f->header->data_hash_table_size) / sizeof(HashItem));
|
2012-06-09 10:30:44 +02:00
|
|
|
p = le64toh(f->data_hash_table[h].tail_hash_offset);
|
2011-10-08 02:20:44 +02:00
|
|
|
if (p == 0) {
|
|
|
|
/* Only entry in the hash table is easy */
|
2011-11-08 18:20:03 +01:00
|
|
|
f->data_hash_table[h].head_hash_offset = htole64(offset);
|
2011-10-08 02:20:44 +02:00
|
|
|
} else {
|
2012-03-06 02:42:32 +01:00
|
|
|
/* Move back to the previous data object, to patch in
|
|
|
|
* pointer */
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
|
2011-10-08 02:20:44 +02:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
o->data.next_hash_offset = htole64(offset);
|
2011-10-08 02:20:44 +02:00
|
|
|
}
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
f->data_hash_table[h].tail_hash_offset = htole64(offset);
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2012-07-16 22:24:02 +02:00
|
|
|
if (JOURNAL_HEADER_CONTAINS(f->header, n_data))
|
|
|
|
f->header->n_data = htole64(le64toh(f->header->n_data) + 1);
|
|
|
|
|
2011-10-08 02:20:44 +02:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
int journal_file_find_data_object_with_hash(
|
|
|
|
JournalFile *f,
|
|
|
|
const void *data, uint64_t size, uint64_t hash,
|
|
|
|
Object **ret, uint64_t *offset) {
|
2012-03-06 02:42:32 +01:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
uint64_t p, osize, h;
|
2011-10-08 02:20:44 +02:00
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(f);
|
|
|
|
assert(data || size == 0);
|
|
|
|
|
|
|
|
osize = offsetof(Object, data.payload) + size;
|
|
|
|
|
2011-12-20 02:38:36 +01:00
|
|
|
if (f->header->data_hash_table_size == 0)
|
|
|
|
return -EBADMSG;
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
h = hash % (le64toh(f->header->data_hash_table_size) / sizeof(HashItem));
|
|
|
|
p = le64toh(f->data_hash_table[h].head_hash_offset);
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
while (p > 0) {
|
|
|
|
Object *o;
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
|
2011-10-08 02:20:44 +02:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
2011-12-21 02:40:59 +01:00
|
|
|
if (le64toh(o->data.hash) != hash)
|
2011-12-27 22:58:20 +01:00
|
|
|
goto next;
|
2011-12-21 02:40:59 +01:00
|
|
|
|
|
|
|
if (o->object.flags & OBJECT_COMPRESSED) {
|
|
|
|
#ifdef HAVE_XZ
|
2012-01-11 22:44:43 +01:00
|
|
|
uint64_t l, rsize;
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-12-21 02:40:59 +01:00
|
|
|
l = le64toh(o->object.size);
|
|
|
|
if (l <= offsetof(Object, data.payload))
|
2011-10-08 02:20:44 +02:00
|
|
|
return -EBADMSG;
|
|
|
|
|
2011-12-21 02:40:59 +01:00
|
|
|
l -= offsetof(Object, data.payload);
|
|
|
|
|
|
|
|
if (!uncompress_blob(o->data.payload, l, &f->compress_buffer, &f->compress_buffer_size, &rsize))
|
|
|
|
return -EBADMSG;
|
|
|
|
|
2012-01-11 22:44:43 +01:00
|
|
|
if (rsize == size &&
|
2011-12-21 02:40:59 +01:00
|
|
|
memcmp(f->compress_buffer, data, size) == 0) {
|
|
|
|
|
|
|
|
if (ret)
|
|
|
|
*ret = o;
|
|
|
|
|
|
|
|
if (offset)
|
|
|
|
*offset = p;
|
|
|
|
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
#else
|
|
|
|
return -EPROTONOSUPPORT;
|
|
|
|
#endif
|
|
|
|
|
|
|
|
} else if (le64toh(o->object.size) == osize &&
|
|
|
|
memcmp(o->data.payload, data, size) == 0) {
|
|
|
|
|
2011-10-08 02:20:44 +02:00
|
|
|
if (ret)
|
|
|
|
*ret = o;
|
|
|
|
|
|
|
|
if (offset)
|
|
|
|
*offset = p;
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
return 1;
|
2011-10-08 02:20:44 +02:00
|
|
|
}
|
|
|
|
|
2011-12-27 22:58:20 +01:00
|
|
|
next:
|
2011-10-08 02:20:44 +02:00
|
|
|
p = le64toh(o->data.next_hash_offset);
|
|
|
|
}
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
int journal_file_find_data_object(
|
|
|
|
JournalFile *f,
|
|
|
|
const void *data, uint64_t size,
|
|
|
|
Object **ret, uint64_t *offset) {
|
|
|
|
|
|
|
|
uint64_t hash;
|
|
|
|
|
|
|
|
assert(f);
|
|
|
|
assert(data || size == 0);
|
|
|
|
|
|
|
|
hash = hash64(data, size);
|
|
|
|
|
|
|
|
return journal_file_find_data_object_with_hash(f,
|
|
|
|
data, size, hash,
|
|
|
|
ret, offset);
|
|
|
|
}
|
|
|
|
|
2012-03-06 02:42:32 +01:00
|
|
|
static int journal_file_append_data(
|
|
|
|
JournalFile *f,
|
|
|
|
const void *data, uint64_t size,
|
|
|
|
Object **ret, uint64_t *offset) {
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
uint64_t hash, p;
|
|
|
|
uint64_t osize;
|
|
|
|
Object *o;
|
|
|
|
int r;
|
2011-12-21 02:40:59 +01:00
|
|
|
bool compressed = false;
|
2011-11-08 18:20:03 +01:00
|
|
|
|
|
|
|
assert(f);
|
|
|
|
assert(data || size == 0);
|
|
|
|
|
|
|
|
hash = hash64(data, size);
|
|
|
|
|
|
|
|
r = journal_file_find_data_object_with_hash(f, data, size, hash, &o, &p);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
else if (r > 0) {
|
|
|
|
|
|
|
|
if (ret)
|
|
|
|
*ret = o;
|
|
|
|
|
|
|
|
if (offset)
|
|
|
|
*offset = p;
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
osize = offsetof(Object, data.payload) + size;
|
|
|
|
r = journal_file_append_object(f, OBJECT_DATA, osize, &o, &p);
|
2011-10-08 02:20:44 +02:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
|
|
|
o->data.hash = htole64(hash);
|
2011-12-21 02:40:59 +01:00
|
|
|
|
|
|
|
#ifdef HAVE_XZ
|
|
|
|
if (f->compress &&
|
|
|
|
size >= COMPRESSION_SIZE_THRESHOLD) {
|
|
|
|
uint64_t rsize;
|
|
|
|
|
|
|
|
compressed = compress_blob(data, size, o->data.payload, &rsize);
|
|
|
|
|
|
|
|
if (compressed) {
|
|
|
|
o->object.size = htole64(offsetof(Object, data.payload) + rsize);
|
|
|
|
o->object.flags |= OBJECT_COMPRESSED;
|
|
|
|
|
|
|
|
f->header->incompatible_flags = htole32(le32toh(f->header->incompatible_flags) | HEADER_INCOMPATIBLE_COMPRESSED);
|
|
|
|
|
|
|
|
log_debug("Compressed data object %lu -> %lu", (unsigned long) size, (unsigned long) rsize);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
#endif
|
|
|
|
|
|
|
|
if (!compressed)
|
|
|
|
memcpy(o->data.payload, data, size);
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
r = journal_file_link_data(f, o, p, hash);
|
2011-10-08 02:20:44 +02:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
2012-03-06 02:42:32 +01:00
|
|
|
/* 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;
|
|
|
|
|
2011-10-08 02:20:44 +02:00
|
|
|
if (ret)
|
|
|
|
*ret = o;
|
|
|
|
|
|
|
|
if (offset)
|
2011-11-08 18:20:03 +01:00
|
|
|
*offset = p;
|
2011-10-08 02:20:44 +02:00
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
uint64_t journal_file_entry_n_items(Object *o) {
|
|
|
|
assert(o);
|
2012-03-01 18:00:01 +01:00
|
|
|
assert(o->object.type == OBJECT_ENTRY);
|
2011-10-08 02:20:44 +02:00
|
|
|
|
|
|
|
return (le64toh(o->object.size) - offsetof(Object, entry.items)) / sizeof(EntryItem);
|
|
|
|
}
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
static uint64_t journal_file_entry_array_n_items(Object *o) {
|
|
|
|
assert(o);
|
2012-03-01 18:00:01 +01:00
|
|
|
assert(o->object.type == OBJECT_ENTRY_ARRAY);
|
2011-11-08 18:20:03 +01:00
|
|
|
|
|
|
|
return (le64toh(o->object.size) - offsetof(Object, entry_array.items)) / sizeof(uint64_t);
|
|
|
|
}
|
|
|
|
|
|
|
|
static int link_entry_into_array(JournalFile *f,
|
2012-03-16 11:59:04 +01:00
|
|
|
le64_t *first,
|
|
|
|
le64_t *idx,
|
2011-11-08 18:20:03 +01:00
|
|
|
uint64_t p) {
|
2011-10-08 02:20:44 +02:00
|
|
|
int r;
|
2011-11-08 18:20:03 +01:00
|
|
|
uint64_t n = 0, ap = 0, q, i, a, hidx;
|
|
|
|
Object *o;
|
|
|
|
|
2011-10-08 02:20:44 +02:00
|
|
|
assert(f);
|
2011-11-08 18:20:03 +01:00
|
|
|
assert(first);
|
|
|
|
assert(idx);
|
|
|
|
assert(p > 0);
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
a = le64toh(*first);
|
|
|
|
i = hidx = le64toh(*idx);
|
|
|
|
while (a > 0) {
|
|
|
|
|
|
|
|
r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &o);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
n = journal_file_entry_array_n_items(o);
|
|
|
|
if (i < n) {
|
|
|
|
o->entry_array.items[i] = htole64(p);
|
|
|
|
*idx = htole64(hidx + 1);
|
|
|
|
return 0;
|
|
|
|
}
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
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 * sizeof(uint64_t),
|
|
|
|
&o, &q);
|
2011-10-08 02:20:44 +02:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
o->entry_array.items[i] = htole64(p);
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
if (ap == 0)
|
2012-03-01 18:00:01 +01:00
|
|
|
*first = htole64(q);
|
2011-10-08 02:20:44 +02:00
|
|
|
else {
|
2011-11-08 18:20:03 +01:00
|
|
|
r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, ap, &o);
|
2011-10-08 02:20:44 +02:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
o->entry_array.next_entry_array_offset = htole64(q);
|
|
|
|
}
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
*idx = htole64(hidx + 1);
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
static int link_entry_into_array_plus_one(JournalFile *f,
|
2012-03-16 11:59:04 +01:00
|
|
|
le64_t *extra,
|
|
|
|
le64_t *first,
|
|
|
|
le64_t *idx,
|
2011-11-08 18:20:03 +01:00
|
|
|
uint64_t p) {
|
|
|
|
|
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(f);
|
|
|
|
assert(extra);
|
|
|
|
assert(first);
|
|
|
|
assert(idx);
|
|
|
|
assert(p > 0);
|
|
|
|
|
|
|
|
if (*idx == 0)
|
|
|
|
*extra = htole64(p);
|
|
|
|
else {
|
2012-03-16 11:59:04 +01:00
|
|
|
le64_t i;
|
2011-11-08 18:20:03 +01:00
|
|
|
|
2012-03-01 18:00:01 +01:00
|
|
|
i = htole64(le64toh(*idx) - 1);
|
2011-11-08 18:20:03 +01:00
|
|
|
r = link_entry_into_array(f, first, &i, p);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
2011-10-08 02:20:44 +02:00
|
|
|
}
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
*idx = htole64(le64toh(*idx) + 1);
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int journal_file_link_entry_item(JournalFile *f, Object *o, uint64_t offset, uint64_t i) {
|
|
|
|
uint64_t p;
|
|
|
|
int r;
|
|
|
|
assert(f);
|
|
|
|
assert(o);
|
|
|
|
assert(offset > 0);
|
|
|
|
|
|
|
|
p = le64toh(o->entry.items[i].object_offset);
|
|
|
|
if (p == 0)
|
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
r = journal_file_move_to_object(f, OBJECT_DATA, p, &o);
|
2011-10-08 02:20:44 +02:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
return link_entry_into_array_plus_one(f,
|
|
|
|
&o->data.entry_offset,
|
|
|
|
&o->data.entry_array_offset,
|
|
|
|
&o->data.n_entries,
|
|
|
|
offset);
|
2011-10-08 02:20:44 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
static int journal_file_link_entry(JournalFile *f, Object *o, uint64_t offset) {
|
2011-11-08 18:20:03 +01:00
|
|
|
uint64_t n, i;
|
2011-10-08 02:20:44 +02:00
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(f);
|
|
|
|
assert(o);
|
|
|
|
assert(offset > 0);
|
2011-11-08 18:20:03 +01:00
|
|
|
assert(o->object.type == OBJECT_ENTRY);
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2012-01-04 02:15:45 +01:00
|
|
|
__sync_synchronize();
|
|
|
|
|
2011-10-08 02:20:44 +02:00
|
|
|
/* Link up the entry itself */
|
2011-11-08 18:20:03 +01:00
|
|
|
r = link_entry_into_array(f,
|
|
|
|
&f->header->entry_array_offset,
|
|
|
|
&f->header->n_entries,
|
|
|
|
offset);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2012-01-07 01:42:13 +01:00
|
|
|
/* log_debug("=> %s seqnr=%lu n_entries=%lu", f->path, (unsigned long) o->entry.seqnum, (unsigned long) f->header->n_entries); */
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
if (f->header->head_entry_realtime == 0)
|
2011-10-13 05:19:35 +02:00
|
|
|
f->header->head_entry_realtime = o->entry.realtime;
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-10-13 05:19:35 +02:00
|
|
|
f->header->tail_entry_realtime = o->entry.realtime;
|
2011-11-08 18:20:03 +01:00
|
|
|
f->header->tail_entry_monotonic = o->entry.monotonic;
|
|
|
|
|
|
|
|
f->tail_entry_monotonic_valid = true;
|
2011-10-08 02:20:44 +02:00
|
|
|
|
|
|
|
/* Link up the items */
|
|
|
|
n = journal_file_entry_n_items(o);
|
|
|
|
for (i = 0; i < n; i++) {
|
|
|
|
r = journal_file_link_entry_item(f, o, offset, i);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int journal_file_append_entry_internal(
|
|
|
|
JournalFile *f,
|
|
|
|
const dual_timestamp *ts,
|
|
|
|
uint64_t xor_hash,
|
|
|
|
const EntryItem items[], unsigned n_items,
|
2011-11-08 18:20:03 +01:00
|
|
|
uint64_t *seqnum,
|
2011-10-08 02:20:44 +02:00
|
|
|
Object **ret, uint64_t *offset) {
|
|
|
|
uint64_t np;
|
|
|
|
uint64_t osize;
|
|
|
|
Object *o;
|
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(f);
|
|
|
|
assert(items || n_items == 0);
|
2011-11-08 18:20:03 +01:00
|
|
|
assert(ts);
|
2011-10-08 02:20:44 +02:00
|
|
|
|
|
|
|
osize = offsetof(Object, entry.items) + (n_items * sizeof(EntryItem));
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
r = journal_file_append_object(f, OBJECT_ENTRY, osize, &o, &np);
|
2011-10-08 02:20:44 +02:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
o->entry.seqnum = htole64(journal_file_seqnum(f, seqnum));
|
2011-10-08 02:20:44 +02:00
|
|
|
memcpy(o->entry.items, items, n_items * sizeof(EntryItem));
|
2011-11-08 18:20:03 +01:00
|
|
|
o->entry.realtime = htole64(ts->realtime);
|
|
|
|
o->entry.monotonic = htole64(ts->monotonic);
|
2011-10-08 02:20:44 +02:00
|
|
|
o->entry.xor_hash = htole64(xor_hash);
|
|
|
|
o->entry.boot_id = f->header->boot_id;
|
|
|
|
|
|
|
|
r = journal_file_link_entry(f, o, np);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
|
|
|
if (ret)
|
|
|
|
*ret = o;
|
|
|
|
|
|
|
|
if (offset)
|
|
|
|
*offset = np;
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2011-12-29 15:00:57 +01:00
|
|
|
void journal_file_post_change(JournalFile *f) {
|
2011-12-19 22:35:46 +01:00
|
|
|
assert(f);
|
|
|
|
|
|
|
|
/* 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. */
|
|
|
|
|
2011-12-20 02:38:36 +01:00
|
|
|
__sync_synchronize();
|
|
|
|
|
2011-12-19 22:35:46 +01:00
|
|
|
if (ftruncate(f->fd, f->last_stat.st_size) < 0)
|
|
|
|
log_error("Failed to to truncate file to its own size: %m");
|
|
|
|
}
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
int journal_file_append_entry(JournalFile *f, const dual_timestamp *ts, const struct iovec iovec[], unsigned n_iovec, uint64_t *seqnum, Object **ret, uint64_t *offset) {
|
2011-10-08 02:20:44 +02:00
|
|
|
unsigned i;
|
|
|
|
EntryItem *items;
|
|
|
|
int r;
|
|
|
|
uint64_t xor_hash = 0;
|
2011-11-08 18:20:03 +01:00
|
|
|
struct dual_timestamp _ts;
|
2011-10-08 02:20:44 +02:00
|
|
|
|
|
|
|
assert(f);
|
|
|
|
assert(iovec || n_iovec == 0);
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
if (!f->writable)
|
|
|
|
return -EPERM;
|
|
|
|
|
|
|
|
if (!ts) {
|
|
|
|
dual_timestamp_get(&_ts);
|
|
|
|
ts = &_ts;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (f->tail_entry_monotonic_valid &&
|
|
|
|
ts->monotonic < le64toh(f->header->tail_entry_monotonic))
|
|
|
|
return -EINVAL;
|
|
|
|
|
2011-12-29 15:00:57 +01:00
|
|
|
items = alloca(sizeof(EntryItem) * n_iovec);
|
2011-10-08 02:20:44 +02:00
|
|
|
|
|
|
|
for (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)
|
2011-12-29 15:00:57 +01:00
|
|
|
return r;
|
2011-10-08 02:20:44 +02:00
|
|
|
|
|
|
|
xor_hash ^= le64toh(o->data.hash);
|
|
|
|
items[i].object_offset = htole64(p);
|
2011-10-15 01:13:37 +02:00
|
|
|
items[i].hash = o->data.hash;
|
2011-10-08 02:20:44 +02:00
|
|
|
}
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
r = journal_file_append_entry_internal(f, ts, xor_hash, items, n_iovec, seqnum, ret, offset);
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-12-19 22:35:46 +01:00
|
|
|
journal_file_post_change(f);
|
|
|
|
|
2011-10-08 02:20:44 +02:00
|
|
|
return r;
|
|
|
|
}
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
static int generic_array_get(JournalFile *f,
|
|
|
|
uint64_t first,
|
|
|
|
uint64_t i,
|
|
|
|
Object **ret, uint64_t *offset) {
|
|
|
|
|
2011-10-08 02:20:44 +02:00
|
|
|
Object *o;
|
2011-12-31 03:55:35 +01:00
|
|
|
uint64_t p = 0, a;
|
2011-10-08 02:20:44 +02:00
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(f);
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
a = first;
|
|
|
|
while (a > 0) {
|
|
|
|
uint64_t n;
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &o);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
n = journal_file_entry_array_n_items(o);
|
|
|
|
if (i < n) {
|
|
|
|
p = le64toh(o->entry_array.items[i]);
|
|
|
|
break;
|
2011-10-08 02:20:44 +02:00
|
|
|
}
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
i -= n;
|
|
|
|
a = le64toh(o->entry_array.next_entry_array_offset);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (a <= 0 || p <= 0)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
|
|
|
if (ret)
|
|
|
|
*ret = o;
|
|
|
|
|
|
|
|
if (offset)
|
|
|
|
*offset = p;
|
|
|
|
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
static int generic_array_get_plus_one(JournalFile *f,
|
|
|
|
uint64_t extra,
|
|
|
|
uint64_t first,
|
|
|
|
uint64_t i,
|
|
|
|
Object **ret, uint64_t *offset) {
|
|
|
|
|
|
|
|
Object *o;
|
|
|
|
|
|
|
|
assert(f);
|
|
|
|
|
|
|
|
if (i == 0) {
|
|
|
|
int r;
|
|
|
|
|
|
|
|
r = journal_file_move_to_object(f, OBJECT_ENTRY, extra, &o);
|
2011-10-08 02:20:44 +02:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
if (ret)
|
|
|
|
*ret = o;
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
if (offset)
|
|
|
|
*offset = extra;
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
return 1;
|
2011-10-08 02:20:44 +02:00
|
|
|
}
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
return generic_array_get(f, first, i-1, ret, offset);
|
|
|
|
}
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
enum {
|
|
|
|
TEST_FOUND,
|
|
|
|
TEST_LEFT,
|
|
|
|
TEST_RIGHT
|
|
|
|
};
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
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,
|
|
|
|
uint64_t *offset,
|
|
|
|
uint64_t *idx) {
|
|
|
|
|
|
|
|
uint64_t a, p, t = 0, i = 0, last_p = 0;
|
|
|
|
bool subtract_one = false;
|
|
|
|
Object *o, *array = NULL;
|
|
|
|
int r;
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
assert(f);
|
|
|
|
assert(test_object);
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
a = first;
|
|
|
|
while (a > 0) {
|
|
|
|
uint64_t left, right, k, lp;
|
|
|
|
|
|
|
|
r = journal_file_move_to_object(f, OBJECT_ENTRY_ARRAY, a, &array);
|
2011-10-08 02:20:44 +02:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
k = journal_file_entry_array_n_items(array);
|
|
|
|
right = MIN(k, n);
|
|
|
|
if (right <= 0)
|
|
|
|
return 0;
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
i = right - 1;
|
|
|
|
lp = p = le64toh(array->entry_array.items[i]);
|
|
|
|
if (p <= 0)
|
|
|
|
return -EBADMSG;
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
r = test_object(f, p, needle);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
if (r == TEST_FOUND)
|
|
|
|
r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
|
|
|
|
|
|
|
|
if (r == TEST_RIGHT) {
|
|
|
|
left = 0;
|
|
|
|
right -= 1;
|
|
|
|
for (;;) {
|
|
|
|
if (left == right) {
|
|
|
|
if (direction == DIRECTION_UP)
|
|
|
|
subtract_one = true;
|
|
|
|
|
|
|
|
i = left;
|
|
|
|
goto found;
|
|
|
|
}
|
|
|
|
|
|
|
|
assert(left < right);
|
|
|
|
|
|
|
|
i = (left + right) / 2;
|
|
|
|
p = le64toh(array->entry_array.items[i]);
|
|
|
|
if (p <= 0)
|
|
|
|
return -EBADMSG;
|
|
|
|
|
|
|
|
r = test_object(f, p, needle);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
if (r == TEST_FOUND)
|
|
|
|
r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
|
|
|
|
|
|
|
|
if (r == TEST_RIGHT)
|
|
|
|
right = i;
|
|
|
|
else
|
|
|
|
left = i + 1;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2012-07-13 00:29:26 +02:00
|
|
|
if (k > n) {
|
|
|
|
if (direction == DIRECTION_UP) {
|
|
|
|
i = n;
|
|
|
|
subtract_one = true;
|
|
|
|
goto found;
|
|
|
|
}
|
|
|
|
|
2011-10-08 02:20:44 +02:00
|
|
|
return 0;
|
2012-07-13 00:29:26 +02:00
|
|
|
}
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
last_p = lp;
|
|
|
|
|
|
|
|
n -= k;
|
|
|
|
t += k;
|
|
|
|
a = le64toh(array->entry_array.next_entry_array_offset);
|
2011-10-08 02:20:44 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
2011-11-08 18:20:03 +01:00
|
|
|
|
|
|
|
found:
|
|
|
|
if (subtract_one && t == 0 && i == 0)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
if (subtract_one && i == 0)
|
|
|
|
p = last_p;
|
|
|
|
else if (subtract_one)
|
|
|
|
p = le64toh(array->entry_array.items[i-1]);
|
|
|
|
else
|
|
|
|
p = le64toh(array->entry_array.items[i]);
|
|
|
|
|
|
|
|
r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
|
|
|
if (ret)
|
|
|
|
*ret = o;
|
|
|
|
|
|
|
|
if (offset)
|
|
|
|
*offset = p;
|
|
|
|
|
|
|
|
if (idx)
|
2012-07-13 00:29:26 +02:00
|
|
|
*idx = t + i + (subtract_one ? -1 : 0);
|
2011-11-08 18:20:03 +01:00
|
|
|
|
|
|
|
return 1;
|
2011-10-08 02:20:44 +02:00
|
|
|
}
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
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,
|
|
|
|
uint64_t *offset,
|
|
|
|
uint64_t *idx) {
|
|
|
|
|
2011-10-08 02:20:44 +02:00
|
|
|
int r;
|
2012-07-13 00:29:26 +02:00
|
|
|
bool step_back = false;
|
|
|
|
Object *o;
|
2011-10-08 02:20:44 +02:00
|
|
|
|
|
|
|
assert(f);
|
2011-11-08 18:20:03 +01:00
|
|
|
assert(test_object);
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
if (n <= 0)
|
|
|
|
return 0;
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
/* This bisects the array in object 'first', but first checks
|
|
|
|
* an extra */
|
|
|
|
r = test_object(f, extra, needle);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
2012-07-12 17:36:04 +02:00
|
|
|
|
|
|
|
if (r == TEST_FOUND)
|
|
|
|
r = direction == DIRECTION_DOWN ? TEST_RIGHT : TEST_LEFT;
|
|
|
|
|
2012-07-13 00:29:26 +02:00
|
|
|
/* 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;
|
2011-11-08 18:20:03 +01:00
|
|
|
|
2012-07-13 00:29:26 +02:00
|
|
|
if (r == TEST_RIGHT) {
|
|
|
|
if (direction == DIRECTION_DOWN)
|
|
|
|
goto found;
|
|
|
|
else
|
|
|
|
return 0;
|
2012-07-12 17:36:04 +02:00
|
|
|
}
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
r = generic_array_bisect(f, first, n-1, needle, test_object, direction, ret, offset, idx);
|
|
|
|
|
2012-07-13 00:29:26 +02:00
|
|
|
if (r == 0 && step_back)
|
|
|
|
goto found;
|
|
|
|
|
2012-07-12 17:36:24 +02:00
|
|
|
if (r > 0 && idx)
|
2011-11-08 18:20:03 +01:00
|
|
|
(*idx) ++;
|
|
|
|
|
|
|
|
return r;
|
2012-07-13 00:29:26 +02:00
|
|
|
|
|
|
|
found:
|
|
|
|
r = journal_file_move_to_object(f, OBJECT_ENTRY, extra, &o);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
|
|
|
if (ret)
|
|
|
|
*ret = o;
|
|
|
|
|
|
|
|
if (offset)
|
|
|
|
*offset = extra;
|
|
|
|
|
|
|
|
if (idx)
|
|
|
|
*idx = 0;
|
|
|
|
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
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,
|
|
|
|
uint64_t *offset) {
|
|
|
|
|
|
|
|
return generic_array_bisect(f,
|
|
|
|
le64toh(f->header->entry_array_offset),
|
|
|
|
le64toh(f->header->n_entries),
|
|
|
|
p,
|
|
|
|
test_object_offset,
|
|
|
|
direction,
|
|
|
|
ret, offset, NULL);
|
2011-11-08 18:20:03 +01:00
|
|
|
}
|
|
|
|
|
2012-07-13 00:29:26 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
static int test_object_seqnum(JournalFile *f, uint64_t p, uint64_t needle) {
|
|
|
|
Object *o;
|
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(f);
|
|
|
|
assert(p > 0);
|
|
|
|
|
|
|
|
r = journal_file_move_to_object(f, OBJECT_ENTRY, p, &o);
|
2011-10-08 02:20:44 +02:00
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
if (le64toh(o->entry.seqnum) == needle)
|
|
|
|
return TEST_FOUND;
|
|
|
|
else if (le64toh(o->entry.seqnum) < needle)
|
|
|
|
return TEST_LEFT;
|
|
|
|
else
|
|
|
|
return TEST_RIGHT;
|
|
|
|
}
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
int journal_file_move_to_entry_by_seqnum(
|
|
|
|
JournalFile *f,
|
|
|
|
uint64_t seqnum,
|
|
|
|
direction_t direction,
|
|
|
|
Object **ret,
|
|
|
|
uint64_t *offset) {
|
|
|
|
|
|
|
|
return generic_array_bisect(f,
|
|
|
|
le64toh(f->header->entry_array_offset),
|
|
|
|
le64toh(f->header->n_entries),
|
|
|
|
seqnum,
|
|
|
|
test_object_seqnum,
|
|
|
|
direction,
|
|
|
|
ret, offset, NULL);
|
|
|
|
}
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
static int test_object_realtime(JournalFile *f, uint64_t p, uint64_t needle) {
|
|
|
|
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;
|
|
|
|
|
|
|
|
if (le64toh(o->entry.realtime) == needle)
|
|
|
|
return TEST_FOUND;
|
|
|
|
else if (le64toh(o->entry.realtime) < needle)
|
|
|
|
return TEST_LEFT;
|
|
|
|
else
|
|
|
|
return TEST_RIGHT;
|
2011-10-08 02:20:44 +02:00
|
|
|
}
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
int journal_file_move_to_entry_by_realtime(
|
|
|
|
JournalFile *f,
|
|
|
|
uint64_t realtime,
|
|
|
|
direction_t direction,
|
|
|
|
Object **ret,
|
|
|
|
uint64_t *offset) {
|
|
|
|
|
|
|
|
return generic_array_bisect(f,
|
|
|
|
le64toh(f->header->entry_array_offset),
|
|
|
|
le64toh(f->header->n_entries),
|
|
|
|
realtime,
|
|
|
|
test_object_realtime,
|
|
|
|
direction,
|
|
|
|
ret, offset, NULL);
|
|
|
|
}
|
|
|
|
|
|
|
|
static int test_object_monotonic(JournalFile *f, uint64_t p, uint64_t needle) {
|
|
|
|
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;
|
|
|
|
|
|
|
|
if (le64toh(o->entry.monotonic) == needle)
|
|
|
|
return TEST_FOUND;
|
|
|
|
else if (le64toh(o->entry.monotonic) < needle)
|
|
|
|
return TEST_LEFT;
|
|
|
|
else
|
|
|
|
return TEST_RIGHT;
|
|
|
|
}
|
|
|
|
|
|
|
|
int journal_file_move_to_entry_by_monotonic(
|
|
|
|
JournalFile *f,
|
|
|
|
sd_id128_t boot_id,
|
|
|
|
uint64_t monotonic,
|
|
|
|
direction_t direction,
|
|
|
|
Object **ret,
|
|
|
|
uint64_t *offset) {
|
|
|
|
|
2012-06-09 15:20:42 +02:00
|
|
|
char t[9+32+1] = "_BOOT_ID=";
|
2011-11-08 18:20:03 +01:00
|
|
|
Object *o;
|
|
|
|
int r;
|
|
|
|
|
2012-07-13 00:29:26 +02:00
|
|
|
assert(f);
|
2011-11-08 18:20:03 +01:00
|
|
|
|
2012-07-13 00:29:26 +02:00
|
|
|
sd_id128_to_string(boot_id, t + 9);
|
2011-11-08 18:20:03 +01:00
|
|
|
r = journal_file_find_data_object(f, t, strlen(t), &o, NULL);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
2012-07-13 00:29:26 +02:00
|
|
|
if (r == 0)
|
2011-11-08 18:20:03 +01:00
|
|
|
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, offset, NULL);
|
|
|
|
}
|
|
|
|
|
|
|
|
int journal_file_next_entry(
|
|
|
|
JournalFile *f,
|
|
|
|
Object *o, uint64_t p,
|
|
|
|
direction_t direction,
|
|
|
|
Object **ret, uint64_t *offset) {
|
|
|
|
|
|
|
|
uint64_t i, n;
|
2011-10-08 02:20:44 +02:00
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(f);
|
2011-11-08 18:20:03 +01:00
|
|
|
assert(p > 0 || !o);
|
|
|
|
|
|
|
|
n = le64toh(f->header->n_entries);
|
|
|
|
if (n <= 0)
|
|
|
|
return 0;
|
2011-10-08 02:20:44 +02:00
|
|
|
|
|
|
|
if (!o)
|
2011-11-08 18:20:03 +01:00
|
|
|
i = direction == DIRECTION_DOWN ? 0 : n - 1;
|
2011-10-08 02:20:44 +02:00
|
|
|
else {
|
2011-11-08 18:20:03 +01:00
|
|
|
if (o->object.type != OBJECT_ENTRY)
|
2011-10-08 02:20:44 +02:00
|
|
|
return -EINVAL;
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
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;
|
|
|
|
|
|
|
|
if (direction == DIRECTION_DOWN) {
|
|
|
|
if (i >= n - 1)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
i++;
|
|
|
|
} else {
|
|
|
|
if (i <= 0)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
i--;
|
|
|
|
}
|
2011-10-08 02:20:44 +02:00
|
|
|
}
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
/* And jump to it */
|
|
|
|
return generic_array_get(f,
|
|
|
|
le64toh(f->header->entry_array_offset),
|
|
|
|
i,
|
|
|
|
ret, offset);
|
|
|
|
}
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
int journal_file_skip_entry(
|
|
|
|
JournalFile *f,
|
|
|
|
Object *o, uint64_t p,
|
|
|
|
int64_t skip,
|
|
|
|
Object **ret, uint64_t *offset) {
|
|
|
|
|
|
|
|
uint64_t i, n;
|
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(f);
|
|
|
|
assert(o);
|
|
|
|
assert(p > 0);
|
|
|
|
|
|
|
|
if (o->object.type != OBJECT_ENTRY)
|
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
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)
|
2011-10-08 02:20:44 +02:00
|
|
|
return r;
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
/* Calculate new index */
|
|
|
|
if (skip < 0) {
|
|
|
|
if ((uint64_t) -skip >= i)
|
|
|
|
i = 0;
|
|
|
|
else
|
|
|
|
i = i - (uint64_t) -skip;
|
|
|
|
} else
|
|
|
|
i += (uint64_t) skip;
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
n = le64toh(f->header->n_entries);
|
|
|
|
if (n <= 0)
|
|
|
|
return -EBADMSG;
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
if (i >= n)
|
|
|
|
i = n-1;
|
|
|
|
|
|
|
|
return generic_array_get(f,
|
|
|
|
le64toh(f->header->entry_array_offset),
|
|
|
|
i,
|
|
|
|
ret, offset);
|
2011-10-08 02:20:44 +02:00
|
|
|
}
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
int journal_file_next_entry_for_data(
|
|
|
|
JournalFile *f,
|
|
|
|
Object *o, uint64_t p,
|
|
|
|
uint64_t data_offset,
|
|
|
|
direction_t direction,
|
|
|
|
Object **ret, uint64_t *offset) {
|
|
|
|
|
|
|
|
uint64_t n, i;
|
2011-10-08 02:20:44 +02:00
|
|
|
int r;
|
2011-11-08 18:20:03 +01:00
|
|
|
Object *d;
|
2011-10-08 02:20:44 +02:00
|
|
|
|
|
|
|
assert(f);
|
2011-11-08 18:20:03 +01:00
|
|
|
assert(p > 0 || !o);
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
|
2011-12-20 00:38:14 +01:00
|
|
|
if (r < 0)
|
2011-11-08 18:20:03 +01:00
|
|
|
return r;
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
n = le64toh(d->data.n_entries);
|
|
|
|
if (n <= 0)
|
|
|
|
return n;
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
if (!o)
|
|
|
|
i = direction == DIRECTION_DOWN ? 0 : n - 1;
|
|
|
|
else {
|
|
|
|
if (o->object.type != OBJECT_ENTRY)
|
|
|
|
return -EINVAL;
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
r = 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_DOWN,
|
|
|
|
NULL, NULL,
|
|
|
|
&i);
|
|
|
|
|
|
|
|
if (r <= 0)
|
2011-10-08 02:20:44 +02:00
|
|
|
return r;
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
if (direction == DIRECTION_DOWN) {
|
|
|
|
if (i >= n - 1)
|
|
|
|
return 0;
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
i++;
|
|
|
|
} else {
|
|
|
|
if (i <= 0)
|
|
|
|
return 0;
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
i--;
|
|
|
|
}
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
}
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
return generic_array_get_plus_one(f,
|
|
|
|
le64toh(d->data.entry_offset),
|
|
|
|
le64toh(d->data.entry_array_offset),
|
|
|
|
i,
|
|
|
|
ret, offset);
|
|
|
|
}
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2012-07-13 00:29:26 +02:00
|
|
|
int journal_file_move_to_entry_by_offset_for_data(
|
|
|
|
JournalFile *f,
|
|
|
|
uint64_t data_offset,
|
|
|
|
uint64_t p,
|
|
|
|
direction_t direction,
|
|
|
|
Object **ret, uint64_t *offset) {
|
|
|
|
|
|
|
|
int r;
|
|
|
|
Object *d;
|
|
|
|
|
|
|
|
assert(f);
|
|
|
|
|
|
|
|
r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
|
|
|
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, offset, NULL);
|
|
|
|
}
|
|
|
|
|
|
|
|
int journal_file_move_to_entry_by_monotonic_for_data(
|
|
|
|
JournalFile *f,
|
|
|
|
uint64_t data_offset,
|
|
|
|
sd_id128_t boot_id,
|
|
|
|
uint64_t monotonic,
|
|
|
|
direction_t direction,
|
|
|
|
Object **ret, uint64_t *offset) {
|
|
|
|
|
|
|
|
char t[9+32+1] = "_BOOT_ID=";
|
|
|
|
Object *o, *d;
|
|
|
|
int r;
|
|
|
|
uint64_t b, z;
|
|
|
|
|
|
|
|
assert(f);
|
|
|
|
|
|
|
|
/* First, seek by time */
|
|
|
|
sd_id128_to_string(boot_id, t + 9);
|
|
|
|
r = journal_file_find_data_object(f, t, strlen(t), &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 */
|
|
|
|
|
|
|
|
for (;;) {
|
|
|
|
Object *qo;
|
|
|
|
uint64_t p, q;
|
|
|
|
|
|
|
|
r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
|
|
|
r = generic_array_bisect_plus_one(f,
|
|
|
|
le64toh(d->data.entry_offset),
|
|
|
|
le64toh(d->data.entry_array_offset),
|
|
|
|
le64toh(d->data.n_entries),
|
|
|
|
z,
|
|
|
|
test_object_offset,
|
|
|
|
direction,
|
|
|
|
NULL, &p, NULL);
|
|
|
|
if (r <= 0)
|
|
|
|
return r;
|
|
|
|
|
|
|
|
r = journal_file_move_to_object(f, OBJECT_DATA, b, &o);
|
|
|
|
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,
|
|
|
|
&qo, &q, NULL);
|
|
|
|
|
|
|
|
if (r <= 0)
|
|
|
|
return r;
|
|
|
|
|
|
|
|
if (p == q) {
|
|
|
|
if (ret)
|
|
|
|
*ret = qo;
|
|
|
|
if (offset)
|
|
|
|
*offset = q;
|
|
|
|
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
z = q;
|
|
|
|
}
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
int journal_file_move_to_entry_by_seqnum_for_data(
|
|
|
|
JournalFile *f,
|
|
|
|
uint64_t data_offset,
|
|
|
|
uint64_t seqnum,
|
|
|
|
direction_t direction,
|
|
|
|
Object **ret, uint64_t *offset) {
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
Object *d;
|
|
|
|
int r;
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2012-07-12 17:36:51 +02:00
|
|
|
assert(f);
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
|
2012-07-12 17:36:51 +02:00
|
|
|
if (r < 0)
|
2011-11-08 18:20:03 +01:00
|
|
|
return r;
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
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, offset, NULL);
|
|
|
|
}
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
int journal_file_move_to_entry_by_realtime_for_data(
|
|
|
|
JournalFile *f,
|
|
|
|
uint64_t data_offset,
|
|
|
|
uint64_t realtime,
|
|
|
|
direction_t direction,
|
|
|
|
Object **ret, uint64_t *offset) {
|
|
|
|
|
|
|
|
Object *d;
|
|
|
|
int r;
|
|
|
|
|
2012-07-12 17:36:51 +02:00
|
|
|
assert(f);
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
r = journal_file_move_to_object(f, OBJECT_DATA, data_offset, &d);
|
2012-07-12 17:36:51 +02:00
|
|
|
if (r < 0)
|
2011-11-08 18:20:03 +01:00
|
|
|
return r;
|
|
|
|
|
|
|
|
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, offset, NULL);
|
2011-10-08 02:20:44 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
void journal_file_dump(JournalFile *f) {
|
|
|
|
Object *o;
|
|
|
|
int r;
|
|
|
|
uint64_t p;
|
|
|
|
|
|
|
|
assert(f);
|
|
|
|
|
2012-07-16 22:24:02 +02:00
|
|
|
journal_file_print_header(f);
|
2011-10-08 02:20:44 +02:00
|
|
|
|
2012-06-05 20:32:55 +02:00
|
|
|
p = le64toh(f->header->header_size);
|
2011-10-08 02:20:44 +02:00
|
|
|
while (p != 0) {
|
2011-11-08 18:20:03 +01:00
|
|
|
r = journal_file_move_to_object(f, -1, p, &o);
|
2011-10-08 02:20:44 +02:00
|
|
|
if (r < 0)
|
|
|
|
goto fail;
|
|
|
|
|
|
|
|
switch (o->object.type) {
|
|
|
|
|
|
|
|
case OBJECT_UNUSED:
|
|
|
|
printf("Type: OBJECT_UNUSED\n");
|
|
|
|
break;
|
|
|
|
|
|
|
|
case OBJECT_DATA:
|
|
|
|
printf("Type: OBJECT_DATA\n");
|
|
|
|
break;
|
|
|
|
|
|
|
|
case OBJECT_ENTRY:
|
2011-10-14 04:44:50 +02:00
|
|
|
printf("Type: OBJECT_ENTRY %llu %llu %llu\n",
|
|
|
|
(unsigned long long) le64toh(o->entry.seqnum),
|
|
|
|
(unsigned long long) le64toh(o->entry.monotonic),
|
|
|
|
(unsigned long long) le64toh(o->entry.realtime));
|
2011-10-08 02:20:44 +02:00
|
|
|
break;
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
case OBJECT_FIELD_HASH_TABLE:
|
|
|
|
printf("Type: OBJECT_FIELD_HASH_TABLE\n");
|
2011-10-08 02:20:44 +02:00
|
|
|
break;
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
case OBJECT_DATA_HASH_TABLE:
|
|
|
|
printf("Type: OBJECT_DATA_HASH_TABLE\n");
|
|
|
|
break;
|
|
|
|
|
|
|
|
case OBJECT_ENTRY_ARRAY:
|
|
|
|
printf("Type: OBJECT_ENTRY_ARRAY\n");
|
2011-10-08 02:20:44 +02:00
|
|
|
break;
|
2012-06-05 20:45:54 +02:00
|
|
|
|
|
|
|
case OBJECT_SIGNATURE:
|
|
|
|
printf("Type: OBJECT_SIGNATURE\n");
|
|
|
|
break;
|
2011-10-08 02:20:44 +02:00
|
|
|
}
|
|
|
|
|
2011-12-21 02:40:59 +01:00
|
|
|
if (o->object.flags & OBJECT_COMPRESSED)
|
|
|
|
printf("Flags: COMPRESSED\n");
|
|
|
|
|
2011-10-08 02:20:44 +02:00
|
|
|
if (p == le64toh(f->header->tail_object_offset))
|
|
|
|
p = 0;
|
|
|
|
else
|
|
|
|
p = p + ALIGN64(le64toh(o->object.size));
|
|
|
|
}
|
|
|
|
|
|
|
|
return;
|
|
|
|
fail:
|
|
|
|
log_error("File corrupt");
|
|
|
|
}
|
|
|
|
|
2012-07-16 22:24:02 +02:00
|
|
|
void journal_file_print_header(JournalFile *f) {
|
|
|
|
char a[33], b[33], c[33];
|
|
|
|
char x[FORMAT_TIMESTAMP_MAX], y[FORMAT_TIMESTAMP_MAX];
|
|
|
|
|
|
|
|
assert(f);
|
|
|
|
|
|
|
|
printf("File Path: %s\n"
|
|
|
|
"File ID: %s\n"
|
|
|
|
"Machine ID: %s\n"
|
|
|
|
"Boot ID: %s\n"
|
|
|
|
"Sequential Number ID: %s\n"
|
2012-07-17 00:52:43 +02:00
|
|
|
"State: %s\n"
|
|
|
|
"Compatible Flags:%s%s\n"
|
|
|
|
"Incompatible Flags:%s%s\n"
|
2012-07-16 22:24:02 +02:00
|
|
|
"Header size: %llu\n"
|
|
|
|
"Arena size: %llu\n"
|
|
|
|
"Data Hash Table Size: %llu\n"
|
|
|
|
"Field Hash Table Size: %llu\n"
|
|
|
|
"Objects: %llu\n"
|
|
|
|
"Entry Objects: %llu\n"
|
|
|
|
"Rotate Suggested: %s\n"
|
|
|
|
"Head Sequential Number: %llu\n"
|
|
|
|
"Tail Sequential Number: %llu\n"
|
|
|
|
"Head Realtime Timestamp: %s\n"
|
|
|
|
"Tail Realtime Timestamp: %s\n",
|
|
|
|
f->path,
|
|
|
|
sd_id128_to_string(f->header->file_id, a),
|
|
|
|
sd_id128_to_string(f->header->machine_id, b),
|
|
|
|
sd_id128_to_string(f->header->boot_id, c),
|
|
|
|
sd_id128_to_string(f->header->seqnum_id, c),
|
2012-07-17 00:52:43 +02:00
|
|
|
f->header->state == STATE_OFFLINE ? "offline" :
|
|
|
|
f->header->state == STATE_ONLINE ? "online" :
|
|
|
|
f->header->state == STATE_ARCHIVED ? "archived" : "unknown",
|
|
|
|
(f->header->compatible_flags & HEADER_COMPATIBLE_SIGNED) ? " SIGNED" : "",
|
|
|
|
(f->header->compatible_flags & ~HEADER_COMPATIBLE_SIGNED) ? " ???" : "",
|
|
|
|
(f->header->incompatible_flags & HEADER_INCOMPATIBLE_COMPRESSED) ? " COMPRESSED" : "",
|
|
|
|
(f->header->incompatible_flags & ~HEADER_INCOMPATIBLE_COMPRESSED) ? " ???" : "",
|
2012-07-16 22:24:02 +02:00
|
|
|
(unsigned long long) le64toh(f->header->header_size),
|
|
|
|
(unsigned long long) le64toh(f->header->arena_size),
|
|
|
|
(unsigned long long) le64toh(f->header->data_hash_table_size) / sizeof(HashItem),
|
|
|
|
(unsigned long long) le64toh(f->header->field_hash_table_size) / sizeof(HashItem),
|
|
|
|
(unsigned long long) le64toh(f->header->n_objects),
|
|
|
|
(unsigned long long) le64toh(f->header->n_entries),
|
|
|
|
yes_no(journal_file_rotate_suggested(f)),
|
|
|
|
(unsigned long long) le64toh(f->header->head_seqnum),
|
|
|
|
(unsigned long long) le64toh(f->header->tail_seqnum),
|
|
|
|
format_timestamp(x, sizeof(x), le64toh(f->header->head_entry_realtime)),
|
|
|
|
format_timestamp(y, sizeof(y), le64toh(f->header->tail_entry_realtime)));
|
|
|
|
|
|
|
|
if (JOURNAL_HEADER_CONTAINS(f->header, n_data))
|
|
|
|
printf("Data Objects: %llu\n"
|
|
|
|
"Data Hash Table Fill: %.1f%%\n",
|
|
|
|
(unsigned long long) 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: %llu\n"
|
|
|
|
"Field Hash Table Fill: %.1f%%\n",
|
|
|
|
(unsigned long long) le64toh(f->header->n_fields),
|
|
|
|
100.0 * (double) le64toh(f->header->n_fields) / ((double) (le64toh(f->header->field_hash_table_size) / sizeof(HashItem))));
|
|
|
|
}
|
|
|
|
|
2011-10-08 02:20:44 +02:00
|
|
|
int journal_file_open(
|
|
|
|
const char *fname,
|
|
|
|
int flags,
|
|
|
|
mode_t mode,
|
2012-07-17 00:36:15 +02:00
|
|
|
JournalMetrics *metrics,
|
2011-10-13 05:19:35 +02:00
|
|
|
JournalFile *template,
|
2011-10-08 02:20:44 +02:00
|
|
|
JournalFile **ret) {
|
|
|
|
|
|
|
|
JournalFile *f;
|
|
|
|
int r;
|
|
|
|
bool newly_created = false;
|
|
|
|
|
|
|
|
assert(fname);
|
|
|
|
|
|
|
|
if ((flags & O_ACCMODE) != O_RDONLY &&
|
|
|
|
(flags & O_ACCMODE) != O_RDWR)
|
|
|
|
return -EINVAL;
|
|
|
|
|
2012-03-15 01:13:01 +01:00
|
|
|
if (!endswith(fname, ".journal"))
|
|
|
|
return -EINVAL;
|
|
|
|
|
2011-10-08 02:20:44 +02:00
|
|
|
f = new0(JournalFile, 1);
|
|
|
|
if (!f)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
2011-10-13 05:19:35 +02:00
|
|
|
f->fd = -1;
|
|
|
|
f->flags = flags;
|
|
|
|
f->mode = mode;
|
2011-10-08 02:20:44 +02:00
|
|
|
f->writable = (flags & O_ACCMODE) != O_RDONLY;
|
|
|
|
f->prot = prot_from_flags(flags);
|
|
|
|
|
2012-07-17 00:36:15 +02:00
|
|
|
if (template)
|
2012-03-15 02:58:27 +01:00
|
|
|
f->compress = template->compress;
|
|
|
|
|
2011-10-08 02:20:44 +02:00
|
|
|
f->path = strdup(fname);
|
|
|
|
if (!f->path) {
|
|
|
|
r = -ENOMEM;
|
|
|
|
goto fail;
|
|
|
|
}
|
|
|
|
|
2011-10-13 05:19:35 +02:00
|
|
|
f->fd = open(f->path, f->flags|O_CLOEXEC, f->mode);
|
|
|
|
if (f->fd < 0) {
|
|
|
|
r = -errno;
|
|
|
|
goto fail;
|
|
|
|
}
|
|
|
|
|
2011-10-08 02:20:44 +02:00
|
|
|
if (fstat(f->fd, &f->last_stat) < 0) {
|
|
|
|
r = -errno;
|
|
|
|
goto fail;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (f->last_stat.st_size == 0 && f->writable) {
|
|
|
|
newly_created = true;
|
|
|
|
|
2011-10-13 05:19:35 +02:00
|
|
|
r = journal_file_init_header(f, template);
|
2011-10-08 02:20:44 +02:00
|
|
|
if (r < 0)
|
|
|
|
goto fail;
|
|
|
|
|
|
|
|
if (fstat(f->fd, &f->last_stat) < 0) {
|
|
|
|
r = -errno;
|
|
|
|
goto fail;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2012-07-16 22:24:02 +02:00
|
|
|
if (f->last_stat.st_size < (off_t) HEADER_SIZE_MIN) {
|
2011-10-08 02:20:44 +02:00
|
|
|
r = -EIO;
|
|
|
|
goto fail;
|
|
|
|
}
|
|
|
|
|
|
|
|
f->header = mmap(NULL, PAGE_ALIGN(sizeof(Header)), prot_from_flags(flags), MAP_SHARED, f->fd, 0);
|
|
|
|
if (f->header == MAP_FAILED) {
|
|
|
|
f->header = NULL;
|
|
|
|
r = -errno;
|
|
|
|
goto fail;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!newly_created) {
|
|
|
|
r = journal_file_verify_header(f);
|
|
|
|
if (r < 0)
|
|
|
|
goto fail;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (f->writable) {
|
2012-07-17 00:36:15 +02:00
|
|
|
if (metrics) {
|
|
|
|
journal_default_metrics(metrics, f->fd);
|
|
|
|
f->metrics = *metrics;
|
|
|
|
} else if (template)
|
|
|
|
f->metrics = template->metrics;
|
|
|
|
|
2011-10-08 02:20:44 +02:00
|
|
|
r = journal_file_refresh_header(f);
|
|
|
|
if (r < 0)
|
|
|
|
goto fail;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (newly_created) {
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
r = journal_file_setup_field_hash_table(f);
|
2011-10-08 02:20:44 +02:00
|
|
|
if (r < 0)
|
|
|
|
goto fail;
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
r = journal_file_setup_data_hash_table(f);
|
2011-10-08 02:20:44 +02:00
|
|
|
if (r < 0)
|
|
|
|
goto fail;
|
|
|
|
}
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
r = journal_file_map_field_hash_table(f);
|
2011-10-08 02:20:44 +02:00
|
|
|
if (r < 0)
|
|
|
|
goto fail;
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
r = journal_file_map_data_hash_table(f);
|
2011-10-08 02:20:44 +02:00
|
|
|
if (r < 0)
|
|
|
|
goto fail;
|
|
|
|
|
|
|
|
if (ret)
|
|
|
|
*ret = f;
|
|
|
|
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
fail:
|
|
|
|
journal_file_close(f);
|
|
|
|
|
|
|
|
return r;
|
|
|
|
}
|
2011-10-13 05:19:35 +02:00
|
|
|
|
|
|
|
int journal_file_rotate(JournalFile **f) {
|
|
|
|
char *p;
|
|
|
|
size_t l;
|
|
|
|
JournalFile *old_file, *new_file = NULL;
|
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(f);
|
|
|
|
assert(*f);
|
|
|
|
|
|
|
|
old_file = *f;
|
|
|
|
|
|
|
|
if (!old_file->writable)
|
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
if (!endswith(old_file->path, ".journal"))
|
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
l = strlen(old_file->path);
|
|
|
|
|
2012-03-15 01:13:01 +01:00
|
|
|
p = new(char, l + 1 + 32 + 1 + 16 + 1 + 16 + 1);
|
2011-10-13 05:19:35 +02:00
|
|
|
if (!p)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
|
|
|
memcpy(p, old_file->path, l - 8);
|
|
|
|
p[l-8] = '@';
|
|
|
|
sd_id128_to_string(old_file->header->seqnum_id, p + l - 8 + 1);
|
|
|
|
snprintf(p + l - 8 + 1 + 32, 1 + 16 + 1 + 16 + 8 + 1,
|
|
|
|
"-%016llx-%016llx.journal",
|
2012-07-16 22:24:02 +02:00
|
|
|
(unsigned long long) le64toh((*f)->header->tail_seqnum),
|
2011-10-13 05:19:35 +02:00
|
|
|
(unsigned long long) le64toh((*f)->header->tail_entry_realtime));
|
|
|
|
|
|
|
|
r = rename(old_file->path, p);
|
|
|
|
free(p);
|
|
|
|
|
|
|
|
if (r < 0)
|
|
|
|
return -errno;
|
|
|
|
|
2012-03-15 17:50:03 +01:00
|
|
|
old_file->header->state = STATE_ARCHIVED;
|
2011-10-13 05:19:35 +02:00
|
|
|
|
2012-07-17 00:36:15 +02:00
|
|
|
r = journal_file_open(old_file->path, old_file->flags, old_file->mode, NULL, old_file, &new_file);
|
2011-10-13 05:19:35 +02:00
|
|
|
journal_file_close(old_file);
|
|
|
|
|
|
|
|
*f = new_file;
|
|
|
|
return r;
|
|
|
|
}
|
|
|
|
|
2012-03-15 01:13:01 +01:00
|
|
|
int journal_file_open_reliably(
|
|
|
|
const char *fname,
|
|
|
|
int flags,
|
|
|
|
mode_t mode,
|
2012-07-17 00:36:15 +02:00
|
|
|
JournalMetrics *metrics,
|
2012-03-15 01:13:01 +01:00
|
|
|
JournalFile *template,
|
|
|
|
JournalFile **ret) {
|
|
|
|
|
|
|
|
int r;
|
|
|
|
size_t l;
|
|
|
|
char *p;
|
|
|
|
|
2012-07-17 00:36:15 +02:00
|
|
|
r = journal_file_open(fname, flags, mode, metrics, template, ret);
|
2012-03-21 23:40:51 +01:00
|
|
|
if (r != -EBADMSG && /* corrupted */
|
|
|
|
r != -ENODATA && /* truncated */
|
|
|
|
r != -EHOSTDOWN && /* other machine */
|
|
|
|
r != -EPROTONOSUPPORT) /* incompatible feature */
|
2012-03-15 01:13:01 +01:00
|
|
|
return r;
|
|
|
|
|
|
|
|
if ((flags & O_ACCMODE) == O_RDONLY)
|
|
|
|
return r;
|
|
|
|
|
|
|
|
if (!(flags & O_CREAT))
|
|
|
|
return r;
|
|
|
|
|
2012-03-15 01:35:03 +01:00
|
|
|
/* The file is corrupted. Rotate it away and try it again (but only once) */
|
|
|
|
|
2012-03-15 01:13:01 +01:00
|
|
|
l = strlen(fname);
|
|
|
|
if (asprintf(&p, "%.*s@%016llx-%016llx.journal~",
|
|
|
|
(int) (l-8), fname,
|
|
|
|
(unsigned long long) now(CLOCK_REALTIME),
|
|
|
|
random_ull()) < 0)
|
|
|
|
return -ENOMEM;
|
|
|
|
|
|
|
|
r = rename(fname, p);
|
|
|
|
free(p);
|
|
|
|
if (r < 0)
|
|
|
|
return -errno;
|
|
|
|
|
|
|
|
log_warning("File %s corrupted, renaming and replacing.", fname);
|
|
|
|
|
2012-07-17 00:36:15 +02:00
|
|
|
return journal_file_open(fname, flags, mode, metrics, template, ret);
|
2012-03-15 01:13:01 +01:00
|
|
|
}
|
|
|
|
|
2011-10-13 05:19:35 +02:00
|
|
|
struct vacuum_info {
|
|
|
|
off_t usage;
|
|
|
|
char *filename;
|
|
|
|
|
|
|
|
uint64_t realtime;
|
|
|
|
sd_id128_t seqnum_id;
|
|
|
|
uint64_t seqnum;
|
2012-03-15 01:35:03 +01:00
|
|
|
|
|
|
|
bool have_seqnum;
|
2011-10-13 05:19:35 +02:00
|
|
|
};
|
|
|
|
|
|
|
|
static int vacuum_compare(const void *_a, const void *_b) {
|
|
|
|
const struct vacuum_info *a, *b;
|
|
|
|
|
|
|
|
a = _a;
|
|
|
|
b = _b;
|
|
|
|
|
2012-03-15 01:35:03 +01:00
|
|
|
if (a->have_seqnum && b->have_seqnum &&
|
|
|
|
sd_id128_equal(a->seqnum_id, b->seqnum_id)) {
|
2011-10-13 05:19:35 +02:00
|
|
|
if (a->seqnum < b->seqnum)
|
|
|
|
return -1;
|
|
|
|
else if (a->seqnum > b->seqnum)
|
|
|
|
return 1;
|
|
|
|
else
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (a->realtime < b->realtime)
|
|
|
|
return -1;
|
|
|
|
else if (a->realtime > b->realtime)
|
|
|
|
return 1;
|
2012-03-15 01:35:03 +01:00
|
|
|
else if (a->have_seqnum && b->have_seqnum)
|
2011-10-13 05:19:35 +02:00
|
|
|
return memcmp(&a->seqnum_id, &b->seqnum_id, 16);
|
2012-03-15 01:35:03 +01:00
|
|
|
else
|
|
|
|
return strcmp(a->filename, b->filename);
|
2011-10-13 05:19:35 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
int journal_directory_vacuum(const char *directory, uint64_t max_use, uint64_t min_free) {
|
|
|
|
DIR *d;
|
|
|
|
int r = 0;
|
|
|
|
struct vacuum_info *list = NULL;
|
|
|
|
unsigned n_list = 0, n_allocated = 0, i;
|
|
|
|
uint64_t sum = 0;
|
|
|
|
|
|
|
|
assert(directory);
|
|
|
|
|
|
|
|
if (max_use <= 0)
|
2011-12-31 02:31:54 +01:00
|
|
|
return 0;
|
2011-10-13 05:19:35 +02:00
|
|
|
|
|
|
|
d = opendir(directory);
|
|
|
|
if (!d)
|
|
|
|
return -errno;
|
|
|
|
|
|
|
|
for (;;) {
|
|
|
|
int k;
|
|
|
|
struct dirent buf, *de;
|
|
|
|
size_t q;
|
|
|
|
struct stat st;
|
|
|
|
char *p;
|
2012-04-13 13:58:50 +02:00
|
|
|
unsigned long long seqnum = 0, realtime;
|
2011-10-13 05:19:35 +02:00
|
|
|
sd_id128_t seqnum_id;
|
2012-03-15 01:35:03 +01:00
|
|
|
bool have_seqnum;
|
2011-10-13 05:19:35 +02:00
|
|
|
|
|
|
|
k = readdir_r(d, &buf, &de);
|
|
|
|
if (k != 0) {
|
|
|
|
r = -k;
|
|
|
|
goto finish;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!de)
|
|
|
|
break;
|
|
|
|
|
2012-03-15 01:35:03 +01:00
|
|
|
if (fstatat(dirfd(d), de->d_name, &st, AT_SYMLINK_NOFOLLOW) < 0)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
if (!S_ISREG(st.st_mode))
|
2011-10-13 05:19:35 +02:00
|
|
|
continue;
|
|
|
|
|
|
|
|
q = strlen(de->d_name);
|
|
|
|
|
2012-03-15 01:35:03 +01:00
|
|
|
if (endswith(de->d_name, ".journal")) {
|
2011-10-13 05:19:35 +02:00
|
|
|
|
2012-03-15 01:35:03 +01:00
|
|
|
/* Vacuum archived files */
|
2011-10-13 05:19:35 +02:00
|
|
|
|
2012-03-15 01:35:03 +01:00
|
|
|
if (q < 1 + 32 + 1 + 16 + 1 + 16 + 8)
|
|
|
|
continue;
|
2011-10-13 05:19:35 +02:00
|
|
|
|
2012-03-15 01:35:03 +01:00
|
|
|
if (de->d_name[q-8-16-1] != '-' ||
|
|
|
|
de->d_name[q-8-16-1-16-1] != '-' ||
|
|
|
|
de->d_name[q-8-16-1-16-1-32-1] != '@')
|
|
|
|
continue;
|
2011-10-13 05:19:35 +02:00
|
|
|
|
2012-03-15 01:35:03 +01:00
|
|
|
p = strdup(de->d_name);
|
|
|
|
if (!p) {
|
|
|
|
r = -ENOMEM;
|
|
|
|
goto finish;
|
|
|
|
}
|
2011-10-13 05:19:35 +02:00
|
|
|
|
2012-03-15 01:35:03 +01:00
|
|
|
de->d_name[q-8-16-1-16-1] = 0;
|
|
|
|
if (sd_id128_from_string(de->d_name + q-8-16-1-16-1-32, &seqnum_id) < 0) {
|
|
|
|
free(p);
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (sscanf(de->d_name + q-8-16-1-16, "%16llx-%16llx.journal", &seqnum, &realtime) != 2) {
|
|
|
|
free(p);
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
|
|
|
have_seqnum = true;
|
|
|
|
|
|
|
|
} else if (endswith(de->d_name, ".journal~")) {
|
|
|
|
unsigned long long tmp;
|
|
|
|
|
|
|
|
/* Vacuum corrupted files */
|
|
|
|
|
|
|
|
if (q < 1 + 16 + 1 + 16 + 8 + 1)
|
|
|
|
continue;
|
2011-10-13 05:19:35 +02:00
|
|
|
|
2012-03-15 01:35:03 +01:00
|
|
|
if (de->d_name[q-1-8-16-1] != '-' ||
|
|
|
|
de->d_name[q-1-8-16-1-16-1] != '@')
|
|
|
|
continue;
|
|
|
|
|
|
|
|
p = strdup(de->d_name);
|
|
|
|
if (!p) {
|
|
|
|
r = -ENOMEM;
|
|
|
|
goto finish;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (sscanf(de->d_name + q-1-8-16-1-16, "%16llx-%16llx.journal~", &realtime, &tmp) != 2) {
|
|
|
|
free(p);
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
|
|
|
have_seqnum = false;
|
|
|
|
} else
|
2011-10-13 05:19:35 +02:00
|
|
|
continue;
|
|
|
|
|
|
|
|
if (n_list >= n_allocated) {
|
|
|
|
struct vacuum_info *j;
|
|
|
|
|
|
|
|
n_allocated = MAX(n_allocated * 2U, 8U);
|
|
|
|
j = realloc(list, n_allocated * sizeof(struct vacuum_info));
|
|
|
|
if (!j) {
|
|
|
|
free(p);
|
|
|
|
r = -ENOMEM;
|
|
|
|
goto finish;
|
|
|
|
}
|
|
|
|
|
|
|
|
list = j;
|
|
|
|
}
|
|
|
|
|
|
|
|
list[n_list].filename = p;
|
2012-03-15 02:59:10 +01:00
|
|
|
list[n_list].usage = 512UL * (uint64_t) st.st_blocks;
|
2011-10-13 05:19:35 +02:00
|
|
|
list[n_list].seqnum = seqnum;
|
|
|
|
list[n_list].realtime = realtime;
|
|
|
|
list[n_list].seqnum_id = seqnum_id;
|
2012-03-15 01:35:03 +01:00
|
|
|
list[n_list].have_seqnum = have_seqnum;
|
2011-10-13 05:19:35 +02:00
|
|
|
|
|
|
|
sum += list[n_list].usage;
|
|
|
|
|
|
|
|
n_list ++;
|
|
|
|
}
|
|
|
|
|
|
|
|
qsort(list, n_list, sizeof(struct vacuum_info), vacuum_compare);
|
|
|
|
|
|
|
|
for(i = 0; i < n_list; i++) {
|
|
|
|
struct statvfs ss;
|
|
|
|
|
|
|
|
if (fstatvfs(dirfd(d), &ss) < 0) {
|
|
|
|
r = -errno;
|
|
|
|
goto finish;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (sum <= max_use &&
|
|
|
|
(uint64_t) ss.f_bavail * (uint64_t) ss.f_bsize >= min_free)
|
|
|
|
break;
|
|
|
|
|
|
|
|
if (unlinkat(dirfd(d), list[i].filename, 0) >= 0) {
|
2012-03-15 02:59:43 +01:00
|
|
|
log_info("Deleted archived journal %s/%s.", directory, list[i].filename);
|
2011-10-13 05:19:35 +02:00
|
|
|
sum -= list[i].usage;
|
|
|
|
} else if (errno != ENOENT)
|
|
|
|
log_warning("Failed to delete %s/%s: %m", directory, list[i].filename);
|
|
|
|
}
|
|
|
|
|
|
|
|
finish:
|
|
|
|
for (i = 0; i < n_list; i++)
|
|
|
|
free(list[i].filename);
|
|
|
|
|
|
|
|
free(list);
|
|
|
|
|
2011-11-08 18:20:03 +01:00
|
|
|
if (d)
|
|
|
|
closedir(d);
|
|
|
|
|
2011-10-13 05:19:35 +02:00
|
|
|
return r;
|
|
|
|
}
|
2011-12-29 15:00:57 +01:00
|
|
|
|
|
|
|
int journal_file_copy_entry(JournalFile *from, JournalFile *to, Object *o, uint64_t p, uint64_t *seqnum, Object **ret, uint64_t *offset) {
|
|
|
|
uint64_t i, n;
|
|
|
|
uint64_t q, xor_hash = 0;
|
|
|
|
int r;
|
|
|
|
EntryItem *items;
|
|
|
|
dual_timestamp ts;
|
|
|
|
|
|
|
|
assert(from);
|
|
|
|
assert(to);
|
|
|
|
assert(o);
|
|
|
|
assert(p);
|
|
|
|
|
|
|
|
if (!to->writable)
|
|
|
|
return -EPERM;
|
|
|
|
|
|
|
|
ts.monotonic = le64toh(o->entry.monotonic);
|
|
|
|
ts.realtime = le64toh(o->entry.realtime);
|
|
|
|
|
|
|
|
if (to->tail_entry_monotonic_valid &&
|
|
|
|
ts.monotonic < le64toh(to->header->tail_entry_monotonic))
|
|
|
|
return -EINVAL;
|
|
|
|
|
|
|
|
n = journal_file_entry_n_items(o);
|
|
|
|
items = alloca(sizeof(EntryItem) * n);
|
|
|
|
|
|
|
|
for (i = 0; i < n; i++) {
|
2012-03-16 11:59:04 +01:00
|
|
|
uint64_t l, h;
|
|
|
|
le64_t le_hash;
|
2011-12-29 15:00:57 +01:00
|
|
|
size_t t;
|
|
|
|
void *data;
|
|
|
|
Object *u;
|
|
|
|
|
|
|
|
q = le64toh(o->entry.items[i].object_offset);
|
|
|
|
le_hash = o->entry.items[i].hash;
|
|
|
|
|
|
|
|
r = journal_file_move_to_object(from, OBJECT_DATA, q, &o);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
|
|
|
if (le_hash != o->data.hash)
|
|
|
|
return -EBADMSG;
|
|
|
|
|
|
|
|
l = le64toh(o->object.size) - offsetof(Object, data.payload);
|
|
|
|
t = (size_t) l;
|
|
|
|
|
|
|
|
/* We hit the limit on 32bit machines */
|
|
|
|
if ((uint64_t) t != l)
|
|
|
|
return -E2BIG;
|
|
|
|
|
|
|
|
if (o->object.flags & OBJECT_COMPRESSED) {
|
|
|
|
#ifdef HAVE_XZ
|
|
|
|
uint64_t rsize;
|
|
|
|
|
|
|
|
if (!uncompress_blob(o->data.payload, l, &from->compress_buffer, &from->compress_buffer_size, &rsize))
|
|
|
|
return -EBADMSG;
|
|
|
|
|
|
|
|
data = from->compress_buffer;
|
|
|
|
l = rsize;
|
|
|
|
#else
|
|
|
|
return -EPROTONOSUPPORT;
|
|
|
|
#endif
|
|
|
|
} else
|
|
|
|
data = o->data.payload;
|
|
|
|
|
|
|
|
r = journal_file_append_data(to, data, l, &u, &h);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
|
|
|
xor_hash ^= le64toh(u->data.hash);
|
|
|
|
items[i].object_offset = htole64(h);
|
|
|
|
items[i].hash = u->data.hash;
|
|
|
|
|
|
|
|
r = journal_file_move_to_object(from, OBJECT_ENTRY, p, &o);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
}
|
|
|
|
|
|
|
|
return journal_file_append_entry_internal(to, &ts, xor_hash, items, n, seqnum, ret, offset);
|
|
|
|
}
|
2011-12-31 02:31:54 +01:00
|
|
|
|
|
|
|
void journal_default_metrics(JournalMetrics *m, int fd) {
|
|
|
|
uint64_t fs_size = 0;
|
|
|
|
struct statvfs ss;
|
2012-01-14 03:06:57 +01:00
|
|
|
char a[FORMAT_BYTES_MAX], b[FORMAT_BYTES_MAX], c[FORMAT_BYTES_MAX], d[FORMAT_BYTES_MAX];
|
2011-12-31 02:31:54 +01:00
|
|
|
|
|
|
|
assert(m);
|
|
|
|
assert(fd >= 0);
|
|
|
|
|
|
|
|
if (fstatvfs(fd, &ss) >= 0)
|
|
|
|
fs_size = ss.f_frsize * ss.f_blocks;
|
|
|
|
|
|
|
|
if (m->max_use == (uint64_t) -1) {
|
|
|
|
|
|
|
|
if (fs_size > 0) {
|
|
|
|
m->max_use = PAGE_ALIGN(fs_size / 10); /* 10% of file system size */
|
|
|
|
|
|
|
|
if (m->max_use > DEFAULT_MAX_USE_UPPER)
|
|
|
|
m->max_use = DEFAULT_MAX_USE_UPPER;
|
|
|
|
|
|
|
|
if (m->max_use < DEFAULT_MAX_USE_LOWER)
|
|
|
|
m->max_use = DEFAULT_MAX_USE_LOWER;
|
|
|
|
} else
|
|
|
|
m->max_use = DEFAULT_MAX_USE_LOWER;
|
|
|
|
} else {
|
|
|
|
m->max_use = PAGE_ALIGN(m->max_use);
|
|
|
|
|
|
|
|
if (m->max_use < JOURNAL_FILE_SIZE_MIN*2)
|
|
|
|
m->max_use = JOURNAL_FILE_SIZE_MIN*2;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (m->max_size == (uint64_t) -1) {
|
|
|
|
m->max_size = PAGE_ALIGN(m->max_use / 8); /* 8 chunks */
|
|
|
|
|
|
|
|
if (m->max_size > DEFAULT_MAX_SIZE_UPPER)
|
|
|
|
m->max_size = DEFAULT_MAX_SIZE_UPPER;
|
|
|
|
} else
|
|
|
|
m->max_size = PAGE_ALIGN(m->max_size);
|
|
|
|
|
|
|
|
if (m->max_size < JOURNAL_FILE_SIZE_MIN)
|
|
|
|
m->max_size = JOURNAL_FILE_SIZE_MIN;
|
|
|
|
|
|
|
|
if (m->max_size*2 > m->max_use)
|
|
|
|
m->max_use = m->max_size*2;
|
|
|
|
|
|
|
|
if (m->min_size == (uint64_t) -1)
|
|
|
|
m->min_size = JOURNAL_FILE_SIZE_MIN;
|
|
|
|
else {
|
|
|
|
m->min_size = PAGE_ALIGN(m->min_size);
|
|
|
|
|
|
|
|
if (m->min_size < JOURNAL_FILE_SIZE_MIN)
|
|
|
|
m->min_size = JOURNAL_FILE_SIZE_MIN;
|
|
|
|
|
|
|
|
if (m->min_size > m->max_size)
|
|
|
|
m->max_size = m->min_size;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (m->keep_free == (uint64_t) -1) {
|
|
|
|
|
|
|
|
if (fs_size > 0) {
|
|
|
|
m->keep_free = PAGE_ALIGN(fs_size / 20); /* 5% of file system size */
|
|
|
|
|
|
|
|
if (m->keep_free > DEFAULT_KEEP_FREE_UPPER)
|
|
|
|
m->keep_free = DEFAULT_KEEP_FREE_UPPER;
|
|
|
|
|
|
|
|
} else
|
|
|
|
m->keep_free = DEFAULT_KEEP_FREE;
|
|
|
|
}
|
|
|
|
|
2012-03-15 02:59:43 +01:00
|
|
|
log_info("Fixed max_use=%s max_size=%s min_size=%s keep_free=%s",
|
|
|
|
format_bytes(a, sizeof(a), m->max_use),
|
|
|
|
format_bytes(b, sizeof(b), m->max_size),
|
|
|
|
format_bytes(c, sizeof(c), m->min_size),
|
|
|
|
format_bytes(d, sizeof(d), m->keep_free));
|
2011-12-31 02:31:54 +01:00
|
|
|
}
|
2012-06-09 10:32:38 +02:00
|
|
|
|
|
|
|
int journal_file_get_cutoff_realtime_usec(JournalFile *f, usec_t *from, usec_t *to) {
|
|
|
|
assert(f);
|
|
|
|
assert(from || to);
|
|
|
|
|
|
|
|
if (from) {
|
2012-07-16 19:28:05 +02:00
|
|
|
if (f->header->head_entry_realtime == 0)
|
|
|
|
return -ENOENT;
|
2012-06-09 10:32:38 +02:00
|
|
|
|
2012-07-16 19:28:05 +02:00
|
|
|
*from = le64toh(f->header->head_entry_realtime);
|
2012-06-09 10:32:38 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
if (to) {
|
2012-07-16 19:28:05 +02:00
|
|
|
if (f->header->tail_entry_realtime == 0)
|
|
|
|
return -ENOENT;
|
2012-06-09 10:32:38 +02:00
|
|
|
|
2012-07-16 19:28:05 +02:00
|
|
|
*to = le64toh(f->header->tail_entry_realtime);
|
2012-06-09 10:32:38 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
|
|
|
|
int journal_file_get_cutoff_monotonic_usec(JournalFile *f, sd_id128_t boot_id, usec_t *from, usec_t *to) {
|
|
|
|
char t[9+32+1] = "_BOOT_ID=";
|
|
|
|
Object *o;
|
|
|
|
uint64_t p;
|
|
|
|
int r;
|
|
|
|
|
|
|
|
assert(f);
|
|
|
|
assert(from || to);
|
|
|
|
|
|
|
|
sd_id128_to_string(boot_id, t + 9);
|
|
|
|
|
|
|
|
r = journal_file_find_data_object(f, t, strlen(t), &o, &p);
|
|
|
|
if (r <= 0)
|
|
|
|
return r;
|
|
|
|
|
|
|
|
if (le64toh(o->data.n_entries) <= 0)
|
|
|
|
return 0;
|
|
|
|
|
|
|
|
if (from) {
|
|
|
|
r = journal_file_move_to_object(f, OBJECT_ENTRY, le64toh(o->data.entry_offset), &o);
|
|
|
|
if (r < 0)
|
|
|
|
return r;
|
|
|
|
|
|
|
|
*from = le64toh(o->entry.monotonic);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (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,
|
|
|
|
&o, NULL);
|
|
|
|
if (r <= 0)
|
|
|
|
return r;
|
|
|
|
|
|
|
|
*to = le64toh(o->entry.monotonic);
|
|
|
|
}
|
|
|
|
|
|
|
|
return 1;
|
|
|
|
}
|
2012-07-16 22:24:02 +02:00
|
|
|
|
|
|
|
bool journal_file_rotate_suggested(JournalFile *f) {
|
|
|
|
assert(f);
|
|
|
|
|
|
|
|
/* If we gained new header fields we gained new features,
|
|
|
|
* hence suggest a rotation */
|
|
|
|
if (le64toh(f->header->header_size) < sizeof(Header))
|
|
|
|
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)
|
|
|
|
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)
|
|
|
|
return true;
|
|
|
|
|
|
|
|
return false;
|
|
|
|
}
|