# HG changeset patch
# User Ian Campbell <ian.campbell@citrix.com>
# Date 1288014999 -3600
# Node ID 3b5c6d7181fecdf6c1043a35047632ddf9950343
# Parent d7eb978c8d5ff37e5982f9ac4799bd662abd4ee1
libxl: Fix migration of HVM guests
In the default "QemuDeviceModelRecord"-style HVM tail of the migration
protocol the size of the qemu save record is unknown at the receiver
and therefore it must read until EOF. This is not compatible with the
xl migration protocol which contains a post-migration handshake and
therefore cannot close the socket on the sending end.
What is required is an explicit length field for the save record,
which the "RemusDeviceModelState"-style HVM tail includes.
Rather than overload the "RemusDeviceModelState" name for a non-Remus
use case (on off chance that they need to diverge for some reason in
the future) introduce a third style called "DeviceModelRecord0002"
which is identical to current "RemusDeviceModelState"-style.
Hopefully the inclusion of a number here will allow easier extension
in the future without needing to come up with increasingly less
helpful names!
Also propagate errors from xc_domain_save and
libxl__domain_suspend_common to callers.
Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
diff -r d7eb978c8d5f -r 3b5c6d7181fe tools/libxc/xc_domain_restore.c
--- a/tools/libxc/xc_domain_restore.c Mon Oct 25 13:54:30 2010 +0100
+++ b/tools/libxc/xc_domain_restore.c Mon Oct 25 14:56:39 2010 +0100
@@ -494,13 +494,18 @@ static int buffer_tail_hvm(xc_interface
return -1;
}
- /* The normal live-migration QEMU record has no length information.
+ /* The legacy live-migration QEMU record has no length information.
* Short of reimplementing the QEMU parser, we''re forced to just
read
- * until EOF. Remus gets around this by sending a different signature
- * which includes a length prefix */
+ * until EOF.
+ *
+ * Gets around this by sending a different signatures for the new
+ * live-migration QEMU record and Remus which includes a length
+ * prefix
+ */
if ( !memcmp(qemusig, "QemuDeviceModelRecord", sizeof(qemusig)) )
return compat_buffer_qemu(xch, ctx, fd, buf);
- else if ( !memcmp(qemusig, "RemusDeviceModelState",
sizeof(qemusig)) )
+ else if ( !memcmp(qemusig, "DeviceModelRecord0002",
sizeof(qemusig)) ||
+ !memcmp(qemusig, "RemusDeviceModelState",
sizeof(qemusig)) )
return buffer_qemu(xch, ctx, fd, buf);
qemusig[20] = ''\0'';
diff -r d7eb978c8d5f -r 3b5c6d7181fe tools/libxc/xg_save_restore.h
--- a/tools/libxc/xg_save_restore.h Mon Oct 25 13:54:30 2010 +0100
+++ b/tools/libxc/xg_save_restore.h Mon Oct 25 14:56:39 2010 +0100
@@ -104,8 +104,9 @@
* Qemu context:
* char[21] : Signature:
* "QemuDeviceModelRecord" : Read Qemu save data until EOF
- * "RemusDeviceModelState" : uint32_t length field followed by
that many
+ * "DeviceModelRecord0002" : uint32_t length field followed by
that many
* bytes of Qemu save data
+ * "RemusDeviceModelState" : Currently the same as
"DeviceModelRecord0002".
*
* PV TAIL:
*
diff -r d7eb978c8d5f -r 3b5c6d7181fe tools/libxl/libxl.c
--- a/tools/libxl/libxl.c Mon Oct 25 13:54:30 2010 +0100
+++ b/tools/libxl/libxl.c Mon Oct 25 14:56:39 2010 +0100
@@ -686,8 +686,8 @@ int libxl_domain_suspend(libxl_ctx *ctx,
int debug = info != NULL && info->flags & XL_SUSPEND_DEBUG;
int rc = 0;
- libxl__domain_suspend_common(ctx, domid, fd, hvm, live, debug);
- if (hvm)
+ rc = libxl__domain_suspend_common(ctx, domid, fd, hvm, live, debug);
+ if (!rc && hvm)
rc = libxl__domain_save_device_model(ctx, domid, fd);
return rc;
}
diff -r d7eb978c8d5f -r 3b5c6d7181fe tools/libxl/libxl_dom.c
--- a/tools/libxl/libxl_dom.c Mon Oct 25 13:54:30 2010 +0100
+++ b/tools/libxl/libxl_dom.c Mon Oct 25 14:56:39 2010 +0100
@@ -442,14 +442,17 @@ int libxl__domain_suspend_common(libxl_c
callbacks.switch_qemu_logdirty =
libxl__domain_suspend_common_switch_qemu_logdirty;
callbacks.data = &si;
- xc_domain_save(ctx->xch, fd, domid, 0, 0, flags, &callbacks, hvm);
+ rc = xc_domain_save(ctx->xch, fd, domid, 0, 0, flags, &callbacks,
hvm);
+ if ( rc ) {
+ LIBXL__LOG_ERRNO(ctx, LIBXL__LOG_ERROR, "saving domain");
+ rc = ERROR_FAIL;
+ }
if (si.suspend_eventchn > 0)
xc_suspend_evtchn_release(ctx->xch, si.xce, domid,
si.suspend_eventchn);
if (si.xce > 0)
xc_evtchn_close(si.xce);
- rc = 0;
out:
libxl__free_all(&gc);
return rc;
@@ -461,15 +464,32 @@ int libxl__domain_save_device_model(libx
int fd2, c;
char buf[1024];
char *filename = libxl__sprintf(&gc,
"/var/lib/xen/qemu-save.%d", domid);
+ struct stat st;
+ uint32_t qemu_state_len;
LIBXL__LOG(ctx, LIBXL__LOG_DEBUG, "Saving device model state to
%s", filename);
libxl__xs_write(&gc, XBT_NULL, libxl__sprintf(&gc,
"/local/domain/0/device-model/%d/command", domid), "save");
libxl__wait_for_device_model(ctx, domid, "paused", NULL, NULL);
+ if (stat(filename, &st) < 0)
+ {
+ LIBXL__LOG(ctx, LIBXL__LOG_ERROR, "Unable to stat qemu save
file\n");
+ return ERROR_FAIL;
+ }
+
+ qemu_state_len = st.st_size;
+ LIBXL__LOG(ctx, LIBXL__LOG_DEBUG, "Qemu state is %d bytes\n",
qemu_state_len);
+
c = libxl_write_exactly(ctx, fd, QEMU_SIGNATURE, strlen(QEMU_SIGNATURE),
"saved-state file", "qemu
signature");
if (c)
return c;
+
+ c = libxl_write_exactly(ctx, fd, &qemu_state_len,
sizeof(qemu_state_len),
+ "saved-state file", "saved-state
length");
+ if (c)
+ return c;
+
fd2 = open(filename, O_RDONLY);
while ((c = read(fd2, buf, sizeof(buf))) != 0) {
if (c < 0) {
diff -r d7eb978c8d5f -r 3b5c6d7181fe tools/libxl/libxl_internal.h
--- a/tools/libxl/libxl_internal.h Mon Oct 25 13:54:30 2010 +0100
+++ b/tools/libxl/libxl_internal.h Mon Oct 25 14:56:39 2010 +0100
@@ -44,7 +44,7 @@
#define LIBXL_PV_EXTRA_MEMORY 1024
#define LIBXL_HVM_EXTRA_MEMORY 2048
#define LIBXL_MIN_DOM0_MEM (128*1024)
-#define QEMU_SIGNATURE "QemuDeviceModelRecord"
+#define QEMU_SIGNATURE "DeviceModelRecord0002"
#define ARRAY_SIZE(a) (sizeof(a) / sizeof(a[0]))
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel
Ian Jackson
2010-Oct-28 11:20 UTC
Re: [Xen-devel] [PATCH] libxl: Fix migration of HVM guests
Ian Campbell writes ("[Xen-devel] [PATCH] libxl: Fix migration of HVM
guests"):> libxl: Fix migration of HVM guests
>
> In the default "QemuDeviceModelRecord"-style HVM tail of the
migration
> protocol the size of the qemu save record is unknown at the receiver
> and therefore it must read until EOF. This is not compatible with the
> xl migration protocol which contains a post-migration handshake and
> therefore cannot close the socket on the sending end.
Thanks, I''ve applied this.
It does point out though that the "DeviceModelRecord0002" token, etc.,
are confused about what layer they are. At some point this should be
done with a new magic record type in xc_domain_{save,restore} and a
new callback ...
Ian.
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel