From: Stefano Stabellini <stefano.stabell...@eu.citrix.com>
We actually need to spawn a second QEMU if we want to run qemu as
non-root, because the pv backends ought to run as root (or device
hotplug may not work).
So in this case, start a second QEMU to provide PV backends in
userspace to HVM guests. Use both dcs->dmss.pvqemu and dcs->dmss.dm
to keep track of the starting QEMUs.
Only proceed when both QEMUs have started.
And, we only default to running QEMU as non-root if we are going to be
able to run split qemus. In particular, it is not safe to run split
qemus if they don't support the emulator_id option, because we need to
split the xenstore paths too.
Signed-off-by: Stefano Stabellini <stefano.stabell...@eu.citrix.com>
Signed-off-by: Ian Jackson <ian.jack...@eu.citrix.com>
---
v6: Split only if trying to run a qemu as non-root
Reorganise changes to dm callbacks
No more dcs in dmss
Use min() to calculate worst rc
Explicitly set unused fields of dmss.pvqemu to 0
Change error handling
v3: use dcs->dmss.pvqemu to spawn the second QEMU
keep track of the rc of both QEMUs before proceeding
---
tools/libxl/libxl_create.c | 72 ++++++++++++++++++++++++++++++++++++++++++++
1 file changed, 72 insertions(+)
diff --git a/tools/libxl/libxl_create.c b/tools/libxl/libxl_create.c
index e67e402..59dfcd67 100644
--- a/tools/libxl/libxl_create.c
+++ b/tools/libxl/libxl_create.c
@@ -431,6 +431,8 @@ static int domcreate_setdefault_dm_user(libxl__gc *gc,
int rc;
const char *user;
+ const char *dm = libxl__domain_device_model(gc, b_info);
+
if (b_info->device_model_user)
/* already set, good-oh */
return 0;
@@ -440,6 +442,15 @@ static int domcreate_setdefault_dm_user(libxl__gc *gc,
/* we're not going to run it anyway */
return 0;
+ if (!libxl__dm_supported(gc, dm, libxl__dm_support_check__emulator_id)) {
+ /* we don't want to run the pv backends as non-root because
+ * device hotplug will no longer work. */
+ LOG(WARN,
+ "Device model does not support split PV backends, running it as root");
+ user = "root";
+ goto found;
+ }
+
user = GCSPRINTF("%s%d", LIBXL_QEMU_USER_BASE, domid);
rc = dm_runas_helper(gc, user);
@@ -802,6 +813,14 @@ static void remus_checkpoint_stream_done(
/* Event callbacks, in this order: */
static void domcreate_dm_support_checked(libxl__egc *egc,
libxl__dm_support_check_state *checking, int rc);
+
+static void domcreate_dm_local_split_pv_cb(libxl__egc *egc,
+ libxl__dm_spawn_state *dmss);
+static void domcreate_dm_local_split_dm_cb(libxl__egc *egc,
+ libxl__dm_spawn_state *dmss);
+static void domcreate_dm_local_split_cb(libxl__egc *egc,
+ libxl__domain_create_state *dcs);
+
static void domcreate_devmodel_started(libxl__egc *egc,
libxl__dm_spawn_state *dmss);
static void domcreate_bootloader_console_available(libxl__egc *egc,
@@ -1044,6 +1063,9 @@ static void domcreate_dm_support_checked(libxl__egc *egc,
/* convenience aliases */
libxl_domain_config *const d_config = dcs->guest_config;
+ const libxl_domain_build_info *b_info = &d_config->b_info;
+ libxl__domain_build_state *const b_state = &dcs->build_state;
+ const uint32_t domid = dcs->guest_domid;
const int restore_fd = dcs->restore_fd;
if (rc) goto out;
@@ -1053,6 +1075,15 @@ static void domcreate_dm_support_checked(libxl__egc *egc,
rc = domcreate_setdefault_dm_user(gc, dcs);
if (rc) goto out;
+ /* run two qemus? */
+ if (d_config->c_info.type == LIBXL_DOMAIN_TYPE_HVM &&
+ !libxl_defbool_val(d_config->b_info.device_model_stubdomain) &&
+ b_info->device_model_user &&
+ strcmp(b_info->device_model_user, "root")) {
+ rc = libxl__dm_emuidmap_add(gc, domid, b_state, EMUID_SPLIT);
+ if (rc) goto out;
+ }
+
dcs->bl.ao = ao;
libxl_device_disk *bootdisk =
d_config->num_disks > 0 ? &d_config->disks[0] : NULL;
@@ -1130,6 +1161,11 @@ static void domcreate_bootloader_done(libxl__egc *egc,
dcs->dmss.dm.callback = domcreate_devmodel_started;
dcs->dmss.callback = domcreate_devmodel_started;
+ dcs->dmss.pvqemu.spawn.ao = ao;
+ dcs->dmss.pvqemu.guest_domid = domid;
+ dcs->dmss.pvqemu.guest_config = 0;
+ dcs->dmss.pvqemu.build_state = 0;
+
if (restore_fd < 0 && dcs->domid_soft_reset == INVALID_DOMID) {
rc = libxl__domain_build(gc, d_config, domid, state);
domcreate_rebuild_done(egc, dcs, rc);
@@ -1398,6 +1434,16 @@ static void domcreate_launch_dm(libxl__egc *egc,
libxl__multidev *multidev,
if (libxl_defbool_val(d_config->b_info.device_model_stubdomain)) {
libxl__spawn_stub_dm(egc, &dcs->dmss, EMUID_DM);
} else {
+ if (state->emuidmap & (1u << EMUID_SPLIT)) {
+ dcs->dmss.dm.rc = 1;
+ dcs->dmss.dm.callback = domcreate_dm_local_split_dm_cb;
+
+ dcs->dmss.pvqemu.rc = 1; /* +ve means in progress */
+ dcs->dmss.pvqemu.callback = domcreate_dm_local_split_pv_cb;
+
+ libxl__spawn_qdisk_backend(egc, &dcs->dmss.pvqemu);
+ }
+
libxl__spawn_local_dm(egc, &dcs->dmss.dm, EMUID_DM);
}
@@ -1455,6 +1501,32 @@ static void domcreate_launch_dm(libxl__egc *egc,
libxl__multidev *multidev,
domcreate_complete(egc, dcs, ret);
}
+static void domcreate_dm_local_split_pv_cb(libxl__egc *egc,
+ libxl__dm_spawn_state *dmss)
+{
+ libxl__domain_create_state *dcs = CONTAINER_OF(dmss, *dcs, dmss.pvqemu);
+ domcreate_dm_local_split_cb(egc, dcs);
+}
+
+static void domcreate_dm_local_split_dm_cb(libxl__egc *egc,
+ libxl__dm_spawn_state *dmss)
+{
+ libxl__domain_create_state *dcs = CONTAINER_OF(dmss, *dcs, dmss.dm);
+ domcreate_dm_local_split_cb(egc, dcs);
+}
+
+static void domcreate_dm_local_split_cb(libxl__egc *egc,
+ libxl__domain_create_state *dcs)
+{
+ if (dcs->dmss.dm.rc > 0 ||
+ dcs->dmss.pvqemu.rc > 0)
+ /* something is still in progress */
+ return;
+
+ dcs->dmss.dm.rc = min(dcs->dmss.dm.rc, dcs->dmss.pvqemu.rc);
+ domcreate_devmodel_started(egc, &dcs->dmss.dm);
+}
+
static void domcreate_devmodel_started(libxl__egc *egc,
libxl__dm_spawn_state *dmss)
{
--
1.7.10.4
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel
