int ploop_copy_init(struct ploop_disk_images_data *di,
struct ploop_copy_param *param,
struct ploop_copy_handle **h)
{
int ret, err;
int blocksize;
char *image = NULL;
char *format = NULL;
char device[64];
char partdev[64];
struct ploop_copy_handle *_h = NULL;
int is_remote;
char mnt[PATH_MAX] = "";
is_remote = is_fd_socket(param->ofd);
if (is_remote < 0) {
ploop_err(0, "Invalid output fd %d: must be a file, "
"a pipe or a socket", param->ofd);
return SYSEXIT_PARAM;
}
if (param->ofd == STDOUT_FILENO)
ploop_set_verbose_level(PLOOP_LOG_NOSTDOUT);
else if (param->ofd == STDERR_FILENO)
ploop_set_verbose_level(PLOOP_LOG_NOCONSOLE);
if (ploop_lock_dd(di))
return SYSEXIT_LOCK;
if (ploop_find_dev_by_dd(di, device, sizeof(device))) {
ploop_err(0, "Can't find running ploop device");
ret = SYSEXIT_SYS;
goto err;
}
ret = get_image_info(device, &image, &format, &blocksize);
if (ret)
goto err;
_h = alloc_ploop_copy_handle(S2B(blocksize));
if (_h == NULL) {
ploop_err(0, "alloc_ploop_copy_handle");
ret = SYSEXIT_MALLOC;
goto err;
}
_h->raw = strcmp(format, "raw") == 0;
_h->ofd = param->ofd;
_h->is_remote = is_remote;
_h->async = param->async;
_h->devfd = open(device, O_RDONLY|O_CLOEXEC);
if (_h->devfd == -1) {
ploop_err(errno, "Can't open device %s", device);
ret = SYSEXIT_DEVICE;
goto err;
}
ret = get_partition_device_name(device, partdev, sizeof(partdev));
if (ret)
goto err;
_h->partfd = open(partdev, O_RDONLY|O_CLOEXEC);
if (_h->partfd == -1) {
ploop_err(errno, "Can't open device %s", partdev);
ret = SYSEXIT_DEVICE;
goto err;
}
ret = SYSEXIT_OPEN;
err = ploop_get_mnt_by_dev(device, mnt, sizeof(mnt));
if (err == -1)
goto err;
else if (err == 0) {
_h->mntfd = open(mnt, O_RDONLY|O_NONBLOCK|O_DIRECTORY);
if (_h->mntfd < 0) {
ploop_err(errno, "Can't open %s", mnt);
goto err;
}
}
ploop_log(0, "Send image %s dev=%s mnt=%s fmt=%s blocksize=%d local=%d",
image, device, mnt, format, blocksize, !is_remote);
if (open_delta(&_h->idelta, image, O_RDONLY|O_DIRECT, OD_ALLOW_DIRTY)) {
ret = SYSEXIT_OPEN;
goto err;
}
ret = complete_running_operation(di, device);
if (ret)
goto err;
_h->cl = register_cleanup_hook(cancel_sender, _h);
pthread_mutex_lock(&_h->sd.wait_mutex);
err:
if (ret) {
ploop_copy_release(_h);
free_ploop_copy_handle(_h);
} else
*h = _h;
free(image);
ploop_unlock_dd(di);
return ret;
}
int ploop_copy_send(struct ploop_copy_send_param *arg)
{
struct delta idelta = { .fd = -1 };
int tracker_on = 0;
int fs_frozen = 0;
int devfd = -1;
int mntfd = -1;
int ret = 0;
char *send_from = NULL;
char *format = NULL;
void *iobuf[2] = {};
int blocksize;
__u64 cluster;
__u64 pos;
__u64 iterpos;
__u64 trackpos;
__u64 trackend;
__u64 xferred;
int iter;
struct ploop_track_extent e;
int i;
pthread_t send_th = 0;
struct send_data sd = {
.mutex = PTHREAD_MUTEX_INITIALIZER,
.cond = PTHREAD_COND_INITIALIZER,
.cond_sent = PTHREAD_COND_INITIALIZER,
};
if (!arg)
return SYSEXIT_PARAM;
sd.fd = arg->ofd;
sd.is_pipe = is_fd_pipe(arg->ofd);
if (sd.is_pipe < 0) {
ploop_err(0, "Invalid output fd %d: must be a file, "
"a pipe or a socket", arg->ofd);
return SYSEXIT_PARAM;
}
if (arg->feedback_fd >= 0 && is_fd_pipe(arg->feedback_fd) != 1) {
ploop_err(errno, "Invalid feedback fd %d: must be "
"a pipe or a socket", arg->feedback_fd);
return SYSEXIT_PARAM;
}
/* If data is to be send to stdout or stderr,
* we have to disable logging to appropriate fd.
*
* As currently there's no way to disable just stderr,
* so in this case we have to disable stdout as well.
*/
if (arg->ofd == STDOUT_FILENO)
ploop_set_verbose_level(PLOOP_LOG_NOSTDOUT);
else if (arg->ofd == STDERR_FILENO)
ploop_set_verbose_level(PLOOP_LOG_NOCONSOLE);
devfd = open(arg->device, O_RDONLY);
if (devfd < 0) {
ploop_err(errno, "Can't open device %s", arg->device);
ret = SYSEXIT_DEVICE;
goto done;
}
mntfd = open_mount_point(arg->device);
if (mntfd < 0) {
/* Error is printed by open_mount_point() */
ret = SYSEXIT_OPEN;
goto done;
}
ret = get_image_info(arg->device, &send_from, &format, &blocksize);
if (ret)
goto done;
cluster = S2B(blocksize);
ret = SYSEXIT_MALLOC;
for (i = 0; i < 2; i++)
if (p_memalign(&iobuf[i], 4096, cluster))
goto done;
ret = complete_running_operation(NULL, arg->device);
if (ret)
goto done;
ret = ioctl_device(devfd, PLOOP_IOC_TRACK_INIT, &e);
if (ret)
goto done;
tracker_on = 1;
if (open_delta_simple(&idelta, send_from, O_RDONLY|O_DIRECT, OD_NOFLAGS)) {
ret = SYSEXIT_OPEN;
goto done;
}
ret = pthread_create(&send_th, NULL, send_thread, &sd);
if (ret) {
ploop_err(ret, "Can't create send thread");
ret = SYSEXIT_SYS;
goto done;
}
ploop_log(-1, "Sending %s", send_from);
trackend = e.end;
for (pos = 0; pos < trackend; ) {
int n;
trackpos = pos + cluster;
ret = ioctl_device(devfd, PLOOP_IOC_TRACK_SETPOS, &trackpos);
if (ret)
goto done;
n = do_pread(cluster, pos);
if (n == 0) /* EOF */
break;
async_send(n, pos);
pos += n;
}
/* First copy done */
iter = 1;
iterpos = 0;
xferred = 0;
for (;;) {
int err;
err = ioctl(devfd, PLOOP_IOC_TRACK_READ, &e);
if (err == 0) {
//fprintf(stderr, "TRACK %llu-%llu\n", e.start, e.end); fflush(stdout);
if (e.end > trackend)
trackend = e.end;
if (e.start < iterpos)
iter++;
iterpos = e.end;
xferred += e.end - e.start;
for (pos = e.start; pos < e.end; ) {
int n;
int copy = e.end - pos;
if (copy > cluster)
copy = cluster;
if (pos + copy > trackpos) {
trackpos = pos + copy;
if (ioctl(devfd, PLOOP_IOC_TRACK_SETPOS, &trackpos)) {
ploop_err(errno, "PLOOP_IOC_TRACK_SETPOS");
ret = SYSEXIT_DEVIOC;
goto done;
}
}
n = do_pread(copy, pos);
if (n == 0) {
ploop_err(0, "Unexpected EOF");
ret = SYSEXIT_READ;
goto done;
}
async_send(n, pos);
pos += n;
}
} else {
if (errno == EAGAIN) /* no more dirty blocks */
break;
ploop_err(errno, "PLOOP_IOC_TRACK_READ");
ret = SYSEXIT_DEVIOC;
goto done;
}
if (iter > 10 || (iter > 1 && xferred > trackend))
break;
}
/* Live iterative transfers are done. Either we transferred
* everything or iterations did not converge. In any case
* now we must suspend VE disk activity. Now it is just
* call of an external program (something sort of
* "killall -9 writetest; sleep 1; umount /mnt2"), actual
* implementation must be intergrated to vzctl/vzmigrate
* and suspend VE with subsequent fsyncing FS.
*/
/* Send the sync command to receiving side. Since older ploop
* might be present on the other side, we need to not break the
* backward compatibility, so just send the first few (SYNC_MARK)
* bytes of delta file contents. New ploop_receive() interprets
* this as "sync me" command, while the old one just writes those
* bytes which is useless but harmless.
*/
if (sd.is_pipe) {
char buf[LEN_STATUS + 1] = {};
ret = do_pread(4096, 0);
if (ret < SYNC_MARK) {
ploop_err(errno, "Short read");
ret = SYSEXIT_READ;
goto done;
}
TS("SEND 0 %d (sync)", SYNC_MARK);
async_send(SYNC_MARK, 0);
/* Now we should wait for the other side to finish syncing
* before freezing the container, to optimize CT frozen time.
*/
if (arg->feedback_fd < 0) {
/* No descriptor to receive a response back is given.
* As ugly as it looks, let's just sleep for some time
* hoping the other side will finish sync.
*/
TS("SLEEP 5");
sleep(5);
goto sync_done;
}
/* Wait for feedback from the receiving side */
/* FIXME: use select/poll with a timeout */
if (read(arg->feedback_fd, buf, LEN_STATUS)
!= LEN_STATUS) {
ploop_err(errno, "Can't read feedback");
ret = SYSEXIT_PROTOCOL;
goto done;
}
if (strncmp(buf, STATUS_OK, LEN_STATUS) == 0) {
goto sync_done;
}
else if (strncmp(buf, STATUS_FAIL, LEN_STATUS) == 0) {
ploop_err(0, "Remote side reported sync failure");
ret = SYSEXIT_FSYNC;
goto done;
}
else {
ploop_err(0, "Got back feedback: %s", buf);
ret = SYSEXIT_PROTOCOL;
goto done;
}
} else {
/* Writing to local file */
fdatasync(arg->ofd);
}
sync_done:
/* Freeze the container */
TS("FLUSH");
ret = run_cmd(arg->flush_cmd);
if (ret)
goto done;
/* Sync fs */
TS("SYNCFS");
if (sys_syncfs(mntfd)) {
ploop_err(errno, "syncfs() failed");
ret = SYSEXIT_FSYNC;
goto done;
}
/* Flush journal and freeze fs (this also clears the fs dirty bit) */
TS("FIFREEZE");
ret = ioctl_device(mntfd, FIFREEZE, 0);
if (ret)
goto done;
fs_frozen = 1;
TS("IOC_SYNC");
ret = ioctl_device(devfd, PLOOP_IOC_SYNC, 0);
if (ret)
goto done;
iter = 1;
iterpos = 0;
for (;;) {
int err;
struct ploop_track_extent e;
err = ioctl(devfd, PLOOP_IOC_TRACK_READ, &e);
if (err == 0) {
__u64 pos;
//fprintf(stderr, "TRACK %llu-%llu\n", e.start, e.end); fflush(stdout);
if (e.end > trackend)
trackend = e.end;
if (e.start < iterpos)
iter++;
iterpos = e.end;
for (pos = e.start; pos < e.end; ) {
int n;
int copy = e.end - pos;
if (copy > cluster)
copy = cluster;
if (pos + copy > trackpos) {
trackpos = pos + copy;
ret = ioctl(devfd, PLOOP_IOC_TRACK_SETPOS, &trackpos);
if (ret)
goto done;
}
TS("READ %llu %d", pos, copy);
n = do_pread(copy, pos);
if (n == 0) {
ploop_err(0, "Unexpected EOF");
ret = SYSEXIT_READ;
goto done;
}
TS("SEND %llu %d", pos, n);
async_send(n, pos);
pos += n;
}
} else {
if (errno == EAGAIN)
break;
ploop_err(errno, "PLOOP_IOC_TRACK_READ");
ret = SYSEXIT_DEVIOC;
goto done;
}
if (iter > 2) {
ploop_err(0, "Too many iterations on frozen FS, aborting");
ret = SYSEXIT_LOOP;
goto done;
}
}
/* Must clear dirty flag on ploop1 image. */
if (strcmp(format, "ploop1") == 0) {
int n;
struct ploop_pvd_header *vh;
TS("READ 0 4096");
n = do_pread(4096, 0);
if (n < SECTOR_SIZE) {
ploop_err(errno, "Short read");
ret = SYSEXIT_READ;
goto done;
}
vh = iobuf[i];
vh->m_DiskInUse = 0;
TS("SEND 0 %d (1st sector)", SECTOR_SIZE);
async_send(SECTOR_SIZE, 0);
}
TS("IOCTL TRACK_STOP");
ret = ioctl(devfd, PLOOP_IOC_TRACK_STOP, 0);
if (ret)
goto done;
tracker_on = 0;
TS("SEND 0 0 (close)");
async_send(0, 0);
pthread_join(send_th, NULL);
send_th = 0;
done:
if (send_th)
pthread_cancel(send_th);
if (fs_frozen)
(void)ioctl_device(mntfd, FITHAW, 0);
if (tracker_on)
(void)ioctl_device(devfd, PLOOP_IOC_TRACK_ABORT, 0);
free(iobuf[0]);
free(iobuf[1]);
if (devfd >=0)
close(devfd);
if (mntfd >=0)
close(mntfd);
free(send_from);
if (idelta.fd >= 0)
close_delta(&idelta);
TS("DONE");
return ret;
}
#undef do_pread
#undef async_send
/* Deprecated, please use ploop_copy_send() instead */
int ploop_send(const char *device, int ofd, const char *flush_cmd,
int is_pipe)
{
struct ploop_copy_send_param s = {
.device = device,
.ofd = ofd,
.flush_cmd = flush_cmd,
};
return ploop_copy_send(&s);
}
static int do_create_snapshot(struct ploop_disk_images_data *di,
const char *guid, const char *snap_dir, int temporary)
{
int ret;
int fd;
char dev[64];
char snap_guid[UUID_SIZE];
char file_guid[UUID_SIZE];
char fname[PATH_MAX];
char conf[PATH_MAX];
char conf_tmp[PATH_MAX];
int online = 0;
int n;
off_t size;
__u32 blocksize;
int version;
if (guid != NULL && !is_valid_guid(guid)) {
ploop_err(0, "Incorrect guid %s", guid);
return SYSEXIT_PARAM;
}
if (is_old_snapshot_format(di))
return SYSEXIT_PARAM;
ret = gen_uuid_pair(snap_guid, sizeof(snap_guid),
file_guid, sizeof(file_guid));
if (ret) {
ploop_err(errno, "Can't generate uuid");
return ret;
}
if (guid != NULL) {
if (find_snapshot_by_guid(di, guid) != -1) {
ploop_err(0, "The snapshot %s already exist",
guid);
return SYSEXIT_PARAM;
}
strcpy(snap_guid, guid);
}
n = get_snapshot_count(di);
if (n == -1) {
return SYSEXIT_PARAM;
} else if (n > 128-2) {
/* The number of images limited by 128
so the snapshot limit 128 - base_image - one_reserverd
*/
ploop_err(errno, "Unable to create a snapshot."
" The maximum number of snapshots (%d) has been reached",
n-1);
return SYSEXIT_PARAM;
}
ret = ploop_find_dev_by_dd(di, dev, sizeof(dev));
if (ret == -1)
return SYSEXIT_SYS;
else if (ret == 0) {
online = 1;
ret = complete_running_operation(di, dev);
if (ret)
return ret;
} else {
ret = get_image_param_offline(di, di->top_guid, &size, &blocksize, &version);
if (ret == SYSEXIT_OPEN && errno == EBUSY) {
/* repair top delta */
char *topdelta[] = {find_image_by_guid(di, di->top_guid), NULL};
blocksize = di->blocksize;
ret = check_deltas(di, topdelta, 0, &blocksize);
if (ret)
return ret;
ret = get_image_param_offline(di, di->top_guid, &size, &blocksize, &version);
}
if (ret)
return ret;
}
ret = merge_temporary_snapshots(di);
if (ret)
return ret;
if (snap_dir != NULL) {
char *name;
char *dir;
dir = realpath(snap_dir, NULL);
if (dir == NULL) {
ploop_err(errno, "Error in realpath(%s)", snap_dir);
return SYSEXIT_CREAT;
}
name = strrchr(di->images[0]->file, '/');
if (name != NULL)
name++;
else
name = di->images[0]->file;
snprintf(fname, sizeof(fname), "%s/%s.%s",
dir, name, file_guid);
free(dir);
} else
snprintf(fname, sizeof(fname), "%s.%s",
di->images[0]->file, file_guid);
ploop_di_change_guid(di, di->top_guid, snap_guid);
if (temporary)
ploop_di_set_temporary(di, snap_guid);
ret = ploop_di_add_image(di, fname, TOPDELTA_UUID, snap_guid);
if (ret)
return ret;
get_disk_descriptor_fname(di, conf, sizeof(conf));
snprintf(conf_tmp, sizeof(conf_tmp), "%s.tmp", conf);
ret = ploop_store_diskdescriptor(conf_tmp, di);
if (ret)
return ret;
if (!online) {
// offline snapshot
fd = create_snapshot_delta(fname, blocksize, size, version);
if (fd < 0) {
ret = SYSEXIT_CREAT;
goto err;
}
close(fd);
} else {
// Always sync fs
ret = create_snapshot(dev, fname, 1);
if (ret)
goto err;
}
if (rename(conf_tmp, conf)) {
ploop_err(errno, "Can't rename %s %s",
conf_tmp, conf);
ret = SYSEXIT_RENAME;
}
if (ret && !online && unlink(fname))
ploop_err(errno, "Can't unlink %s",
fname);
ploop_log(0, "ploop %s %s has been successfully created",
get_snap_str(temporary), snap_guid);
err:
if (ret && unlink(conf_tmp))
ploop_err(errno, "Can't unlink %s", conf_tmp);
return ret;
}
int do_create_snapshot(struct ploop_disk_images_data *di,
const char *guid, const char *snap_dir,
const char *cbt_uuid, int flags)
{
int ret, rc;
int fd;
char dev[64];
char snap_guid[UUID_SIZE];
char top_guid[UUID_SIZE];
char file_guid[UUID_SIZE];
char fname[PATH_MAX];
const char *prev_fname = NULL;
char conf[PATH_MAX];
char conf_tmp[PATH_MAX];
int online = 0;
int temporary = flags & SNAP_TYPE_TEMPORARY;
int n;
off_t size;
__u32 blocksize;
int version;
uuid_t u;
const __u8 *cbt_u = NULL;
if (cbt_uuid != NULL) {
ploop_log(0, "Create snapshot CBT uuid=%s", cbt_uuid);
if (uuid_parse(cbt_uuid, u)) {
ploop_log(-1, "Incorrect cbt uuid is specified %s",
cbt_uuid);
return SYSEXIT_PARAM;
}
cbt_u = u;
}
if (guid != NULL && !is_valid_guid(guid)) {
ploop_err(0, "Incorrect guid %s", guid);
return SYSEXIT_PARAM;
}
if (is_old_snapshot_format(di))
return SYSEXIT_PARAM;
ret = gen_uuid_pair(snap_guid, sizeof(snap_guid),
file_guid, sizeof(file_guid));
if (ret)
return ret;
if (di->vol && di->vol->parent) {
ret = ploop_uuid_generate(top_guid, sizeof(top_guid));
if (ret)
return ret;
} else
strcpy(top_guid, TOPDELTA_UUID);
if (guid != NULL) {
if (find_snapshot_by_guid(di, guid) != -1) {
ploop_err(0, "The snapshot %s already exist",
guid);
return SYSEXIT_PARAM;
}
strcpy(snap_guid, guid);
}
n = get_snapshot_count(di);
if (n == -1) {
return SYSEXIT_PARAM;
} else if (n > 128-2) {
/* The number of images limited by 128
so the snapshot limit 128 - base_image - one_reserverd
*/
ploop_err(errno, "Unable to create a snapshot."
" The maximum number of snapshots (%d) has been reached",
n-1);
return SYSEXIT_PARAM;
}
rc = ploop_find_dev_by_dd(di, dev, sizeof(dev));
if (rc == -1)
return SYSEXIT_SYS;
if (rc == 0) {
if (flags & SNAP_TYPE_OFFLINE) {
ret = get_image_param_online(dev, &size,
&blocksize, &version);
} else {
online = 1;
ret = complete_running_operation(di, dev);
}
if (ret)
return ret;
} else {
ret = get_image_param_offline(di, di->top_guid, &size, &blocksize, &version);
if (ret == SYSEXIT_OPEN && errno == EBUSY) {
/* repair top delta */
char *topdelta[] = {find_image_by_guid(di, di->top_guid), NULL};
blocksize = di->blocksize;
ret = check_deltas(di, topdelta, 0, &blocksize, NULL);
if (ret)
return ret;
ret = get_image_param_offline(di, di->top_guid, &size, &blocksize, &version);
}
if (ret)
return ret;
}
ret = merge_temporary_snapshots(di);
if (ret)
return ret;
if (snap_dir != NULL) {
char *name;
char *dir;
dir = realpath(snap_dir, NULL);
if (dir == NULL) {
ploop_err(errno, "Error in realpath(%s)", snap_dir);
return SYSEXIT_CREAT;
}
name = strrchr(di->images[0]->file, '/');
if (name != NULL)
name++;
else
name = di->images[0]->file;
snprintf(fname, sizeof(fname), "%s/%s.%s",
dir, name, file_guid);
free(dir);
} else
snprintf(fname, sizeof(fname), "%s.%s",
di->images[0]->file, file_guid);
prev_fname = find_image_by_guid(di, di->top_guid);
if (prev_fname == NULL) {
ploop_err(0, "Unable to find image by uuid %s",
di->top_guid);
return SYSEXIT_PARAM;
}
ploop_di_change_guid(di, di->top_guid, snap_guid);
if (temporary)
ploop_di_set_temporary(di, snap_guid);
ret = ploop_di_add_image(di, fname, top_guid, snap_guid);
if (ret)
return ret;
get_disk_descriptor_fname(di, conf, sizeof(conf));
snprintf(conf_tmp, sizeof(conf_tmp), "%s.tmp", conf);
ret = ploop_store_diskdescriptor(conf_tmp, di);
if (ret)
return ret;
if (!online) {
// offline snapshot
ret = 0;
fd = create_snapshot_delta(fname, blocksize, size, version);
if (fd < 0) {
ret = SYSEXIT_CREAT;
goto err;
}
close(fd);
if (cbt_u != NULL)
ret = write_empty_cbt_to_image(fname, prev_fname, cbt_u);
else if (di->mode != PLOOP_RAW_MODE) {
if (rc == 0)
ret = cbt_dump(di, dev, fname);
else
ret = ploop_move_cbt(fname, prev_fname);
}
if (ret)
goto err;
} else {
// Always sync fs
ret = create_snapshot(dev, fname, 1, cbt_u, prev_fname);
if (ret)
goto err;
}
if (rename(conf_tmp, conf)) {
ploop_err(errno, "Can't rename %s %s",
conf_tmp, conf);
ret = SYSEXIT_RENAME;
}
if (ret && !online && unlink(fname))
ploop_err(errno, "Can't unlink %s",
fname);
ploop_log(0, "ploop %s %s has been successfully created",
get_snap_str(temporary), snap_guid);
err:
if (ret && unlink(conf_tmp))
ploop_err(errno, "Can't unlink %s", conf_tmp);
return ret;
}
