static struct ploop_copy_handle *alloc_ploop_copy_handle(int cluster)
{
struct ploop_copy_handle *h;
h = calloc(1, sizeof(struct ploop_copy_handle));
if (h == NULL)
return NULL;
pthread_mutex_init(&h->sd.mutex, NULL);
pthread_cond_init(&h->sd.cond, NULL);
pthread_mutex_init(&h->sd.wait_mutex, NULL);
pthread_cond_init(&h->sd.wait_cond, NULL);
pthread_barrier_init(&h->sd.barrier, NULL, 2);
h->devfd = h->ofd = h->mntfd = h->idelta.fd = -1;
h->cluster = cluster;
if (p_memalign(&h->iobuf[0], 4096, cluster))
goto err;
if (p_memalign(&h->iobuf[1], 4096, cluster))
goto err;
return h;
err:
free_ploop_copy_handle(h);
return NULL;
}
static int send_optional_header(struct ploop_copy_handle *copy_h)
{
int ret;
struct ploop_pvd_header *vh;
size_t block_size;
struct ploop_pvd_ext_block_check *hc;
struct ploop_pvd_ext_block_element_header *h;
__u8 *block = NULL, *data;
vh = (struct ploop_pvd_header *)copy_h->idelta.hdr0;
block_size = vh->m_Sectors * SECTOR_SIZE;
if (p_memalign((void **)&block, 4096, block_size))
return SYSEXIT_MALLOC;
hc = (struct ploop_pvd_ext_block_check *)block;
h = (struct ploop_pvd_ext_block_element_header *)(hc + 1);
data = (__u8 *)(h + 1);
h->magic = EXT_MAGIC_DIRTY_BITMAP;
ret = save_dirty_bitmap(copy_h->devfd, ©_h->idelta,
copy_h->eof_offset, data, &h->size,
NULL, cbt_writer, copy_h);
if (ret) {
if (ret == SYSEXIT_NOCBT)
ret = 0;
goto out;
}
vh->m_DiskInUse = SIGNATURE_DISK_CLOSED_V21;
vh->m_FormatExtensionOffset = (copy_h->eof_offset + SECTOR_SIZE - 1) / SECTOR_SIZE;
if (send_buf(copy_h, vh, sizeof(*vh), 0)) {
ploop_err(errno, "Can't write header");
ret = SYSEXIT_WRITE;
goto out;
}
ploop_log(3, "Send extension header offset=%llu size=%d",
vh->m_FormatExtensionOffset * SECTOR_SIZE, h->size);
hc->m_Magic = FORMAT_EXTENSION_MAGIC;
MD5((const unsigned char *)(hc + 1), block_size - sizeof(*hc), hc->m_Md5);
if (send_buf(copy_h, block, block_size, vh->m_FormatExtensionOffset * SECTOR_SIZE)) {
ploop_err(errno, "Can't write optional header");
ret = SYSEXIT_WRITE;
goto out;
}
out:
free(block);
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);
}
int ploop_copy_receiver(struct ploop_copy_receive_param *arg)
{
int ofd, ret;
__u64 cluster = 0;
void *iobuf = NULL;
int n;
struct pcopy_pkt_desc desc;
if (!arg)
return SYSEXIT_PARAM;
if (is_fd_socket(arg->ifd) != 1) {
ploop_err(errno, "Invalid input fd %d: must be "
"a pipe or a socket", arg->ifd);
return SYSEXIT_PARAM;
}
ofd = open(arg->file, O_WRONLY|O_CREAT|O_TRUNC, 0600);
if (ofd < 0) {
ploop_err(errno, "Can't open %s", arg->file);
return SYSEXIT_CREAT;
}
ploop_dbg(3, "RCV start %s", arg->file);
for (;;) {
if (nread(arg->ifd, &desc, sizeof(desc)) < 0) {
ploop_err(errno, "Error in nread(desc)");
ret = SYSEXIT_READ;
goto out;
}
if (desc.marker != PCOPY_MARKER) {
ploop_err(0, "Stream corrupted");
ret = SYSEXIT_PROTOCOL;
goto out;
}
if (desc.size > cluster) {
free(iobuf);
iobuf = NULL;
cluster = desc.size;
if (p_memalign(&iobuf, 4096, cluster)) {
ret = SYSEXIT_MALLOC;
goto out;
}
}
if (desc.size == 0)
break;
if (nread(arg->ifd, iobuf, desc.size)) {
ploop_err(errno, "Error in nread data");
ret = SYSEXIT_READ;
goto out;
}
ploop_log(3, "RCV type=%d len=%d pos=%" PRIu64,
desc.type, desc.size, (uint64_t)desc.pos);
ret = 0;
switch (desc.type) {
case PCOPY_PKT_DATA:
case PCOPY_PKT_DATA_ASYNC: {
n = TEMP_FAILURE_RETRY(pwrite(ofd, iobuf, desc.size, desc.pos));
if (n != desc.size) {
if (n < 0)
ploop_err(errno, "Error in pwrite");
else
ploop_err(0, "Error: short pwrite");
ret = SYSEXIT_WRITE;
goto out;
}
break;
}
case PCOPY_PKT_CMD: {
unsigned int cmd = ((unsigned int *) iobuf)[0];
switch(cmd) {
case PCOPY_CMD_SYNC:
ret = data_sync(ofd);
if (ret)
goto out;
break;
default:
ploop_err(0, "ploop_copy_receiver: unsupported command %d",
cmd);
ret = SYSEXIT_PARAM;
}
break;
}
default:
ploop_err(0, "ploop_copy_receiver: unsupported command type%d",
desc.type);
ret = SYSEXIT_PARAM;
break;
}
/* send reply */
if (desc.type != PCOPY_PKT_DATA_ASYNC &&
nwrite(arg->ifd, &ret, sizeof(int))) {
ret = SYSEXIT_WRITE;
ploop_err(errno, "failed to send reply");
goto out;
}
}
ret = data_sync(ofd);
if (ret)
goto out;
ploop_dbg(3, "RCV exited");
/* send final reply */
ret = 0;
if (nwrite(arg->ifd, &ret, sizeof(int))) {
ret = SYSEXIT_WRITE;
ploop_err(errno, "failed to send reply");
goto out;
}
out:
if (close(ofd)) {
ploop_err(errno, "Error in close");
if (!ret)
ret = SYSEXIT_WRITE;
}
if (ret)
unlink(arg->file);
free(iobuf);
return ret;
}
int ploop_copy_receive(struct ploop_copy_receive_param *arg)
{
int ofd, ret;
__u64 cluster = 0;
void *iobuf = NULL;
if (!arg)
return SYSEXIT_PARAM;
if (is_fd_pipe(arg->ifd) != 1) {
ploop_err(errno, "Invalid input fd %d: must be "
"a pipe or a socket", arg->ifd);
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 feedback 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->feedback_fd == STDOUT_FILENO)
ploop_set_verbose_level(PLOOP_LOG_NOSTDOUT);
else if (arg->feedback_fd == STDERR_FILENO)
ploop_set_verbose_level(PLOOP_LOG_NOCONSOLE);
ofd = open(arg->file, O_WRONLY|O_CREAT|O_EXCL, 0600);
if (ofd < 0) {
ploop_err(errno, "Can't open %s", arg->file);
return SYSEXIT_CREAT;
}
/* Read data */
for (;;) {
int n;
struct xfer_desc desc;
if (nread(arg->ifd, &desc, sizeof(desc)) < 0) {
ploop_err(errno, "Error in nread(desc)");
ret = SYSEXIT_READ;
goto out;
}
if (desc.marker != PLOOPCOPY_MARKER) {
ploop_err(0, "Stream corrupted");
ret = SYSEXIT_PROTOCOL;
goto out;
}
if (desc.size > cluster) {
free(iobuf);
iobuf = NULL;
cluster = desc.size;
if (p_memalign(&iobuf, 4096, cluster)) {
ret = SYSEXIT_MALLOC;
goto out;
}
}
if (desc.size == 0)
break;
if (nread(arg->ifd, iobuf, desc.size)) {
ploop_err(errno, "Error in nread data");
ret = SYSEXIT_READ;
goto out;
}
if (desc.size == SYNC_MARK) {
int st;
/* ignore received data, instead do sync */
st = fdatasync(ofd);
if (arg->feedback_fd >= 0) {
/* Tell the sending side how it went */
int w;
w = write(arg->feedback_fd,
st ? STATUS_FAIL : STATUS_OK,
LEN_STATUS);
/* check write error only if no error yet */
if (!st && w != LEN_STATUS) {
ploop_err(errno, "Error in write(%d)",
arg->feedback_fd);
ret = SYSEXIT_WRITE;
goto out;
}
}
if (st) {
ploop_err(errno, "Error in fdatasync()");
ret = SYSEXIT_WRITE;
goto out;
}
continue;
}
n = pwrite(ofd, iobuf, desc.size, desc.pos);
if (n != desc.size) {
if (n < 0)
ploop_err(errno, "Error in pwrite");
else
ploop_err(0, "Error: short pwrite");
ret = SYSEXIT_WRITE;
goto out;
}
}
if (fdatasync(ofd)) {
ploop_err(errno, "Error in fdatasync");
ret = SYSEXIT_WRITE;
goto out;
}
ret = 0;
out:
if (close(ofd)) {
ploop_err(errno, "Error in close");
if (!ret)
ret = SYSEXIT_WRITE;
}
if (ret)
unlink(arg->file);
free(iobuf);
return ret;
}
