altcloud/fence-virt
3
0
Fork 1
Code Issues Pull Requests Releases Activity

fence_virtd: Remove checkpoint, replace it with a CPG only plugin

Browse Source 
Remove the checkpoint backend plugin, as it used CMAN and openais
checkpoints. Given nobody is using CMAN or openais anymore, the plugin
was useless for clusters running the current stack.

Signed-off-by: Ryan McCabe <rmccabe@redhat.com>
This commit is contained in:
Ryan McCabe 2017-05-27 21:19:41 -04:00
parent 3bfe80d293
commit 7f6798916a
14 changed files with 281 additions and 1526 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

21
README
View File

@ -42,15 +42,14 @@ has failed.
In order to be able to guarantee safe fencing of a VM even if the
last- known host is down, we must store the last-known locations of
each virtual machine in some sort of cluster-wide way. For this, we
use the AIS Checkpointing API, which is provided by OpenAIS. Every
few seconds, fence_virtd queries the hypervisor via libvirt and
stores any local VM states in a checkpoint. In the event of a
physical node failure (which consequently causes the failure of one
or more guests), we can then read the checkpoint section corresponding
to the guest we need to fence to find out the previous owner. With
that information, we can then check with CMAN to see if the last-
known host node has been fenced. If so, then the VM is clean as well.
The physical cluster must, therefore, have fencing in order for
use the corosync CPG API. Every few seconds, fence_virtd queries the
hypervisor via libvirt and stores any local VM states and sends those
states over CPG to all other members. In the event of a physical node
failure (which consequently causes the failure of one or more guests),
we can then read the stored VM state corresponding to the guest we need
to fence to find out the previous owner. With that information, we can
infer if the known host node has been fenced. If so, then the VM is clean
as well. The physical cluster must, therefore, have fencing in order for
fence_virtd to work.
Operation of a node hosting a VM which needs to be fenced:
@ -68,7 +67,7 @@ Operation of high-node-ID:
(a) Receive multicast packet
(b) Authenticate multicast packet
(c) Read VM state from checkpoint
(c) Read VM state from stored CPG messages
(d) Check liveliness of nodeID hosting VM (if alive, do nothing)
(e) Open connection to host contained within multicast
packet.
@ -77,7 +76,7 @@ Operation of high-node-ID:
(h) Authenticate server & send response.
NOTE: There is always a possibility that a VM is started again
before the fencing operation and checkpoint update for that VM
before the fencing operation and CPG update for that VM
occurs. If the VM has booted and rejoined the cluster, fencing will
not be necessary. If it is in the process of booting, but has not
yet joined the cluster, fencing will also not be necessary - because

17
configure.in
View File

@ -2,7 +2,7 @@
# Process this file with autoconf to produce a configure script.
AC_PREREQ(2.59)
AC_INIT(fence-virt, 0.3.3, rmccabe@redhat.com)
AC_INIT(fence-virt, 0.4.0, rmccabe@redhat.com)
AC_CONFIG_SRCDIR([server/main.c])
AC_CONFIG_HEADER([include/config.h])
@ -18,7 +18,6 @@ AC_PROG_LN_S
AC_PROG_MAKE_SET
# Checks for libraries.
AC_CHECK_LIB([SaCkpt], [saCkptCheckpointOpen])
AC_CHECK_LIB([cpg], [cpg_finalize])
AC_CHECK_LIB([nss3], [NSS_NoDB_Init])
AC_CHECK_LIB([pthread], [pthread_create])
@ -33,7 +32,7 @@ AC_SUBST(pm_ver)
AC_HEADER_DIRENT
AC_HEADER_STDC
AC_HEADER_SYS_WAIT
AC_CHECK_HEADERS([arpa/inet.h fcntl.h malloc.h netdb.h netinet/in.h stdint.h stdlib.h string.h sys/ioctl.h sys/socket.h sys/time.h syslog.h termios.h unistd.h openais/cpg.h corosync/cpg.h])
AC_CHECK_HEADERS([arpa/inet.h fcntl.h malloc.h netdb.h netinet/in.h stdint.h stdlib.h string.h sys/ioctl.h sys/socket.h sys/time.h syslog.h termios.h unistd.h corosync/cpg.h])
# Checks for typedefs, structures, and compiler characteristics.
AC_C_CONST
@ -80,12 +79,12 @@ AC_ARG_ENABLE(libvirt-plugin,
[ mod_libvirt=$enableval ], [ mod_libvirt=yes ])
AC_SUBST(mod_libvirt)
# checkpoint plugin: Disabled by default
AC_ARG_ENABLE(checkpoint-plugin,
[AS_HELP_STRING([--enable-checkpoint-plugin],
[Enable CMAN / AIS checkpoint backend plugin])],
[ mod_checkpoint=$enableval ], [ mod_checkpoint=no ])
AC_SUBST(mod_checkpoint)
# cpg plugin: Disabled by default
AC_ARG_ENABLE(cpg-plugin,
[AS_HELP_STRING([--enable-cpg-plugin],
[Enable CPG/libvirt backend plugin])],
[ mod_cpg=$enableval ], [ mod_cpg=no ])
AC_SUBST(mod_cpg)
# libvirt-qmf plugin: Disabled by default
AC_ARG_ENABLE(libvirt-qmf-plugin,

8
fence_virt.txt
View File

@ -80,13 +80,13 @@ We propose at 5 plugins in this case:
* Libvirt (local-only). There is no intracommunication and no
migration support is provided
* Cluster checkpoint (+ libvirt). This the way fence_xvmd
* Cluster CPG (+ libvirt). This the way fence_xvmd
operates today. This setup has the most requirements on the
infrastructure, as it requires guest to host networking _and_
host-to-host clustering in order to keep track of virtual
machines. The benefit is that it is self-contained and requires
no external management nodes. VM states are stored in checkpoints
so that other hosts know the locations of other VMs and can make
no external management nodes. VM states are stored so that other
CPG group members know the locations of other VMs and can make
some decisions about whether a VM is dead based on whether a host
is dead (i.e. if fencing is in use or can be performed on the
host).
@ -113,7 +113,7 @@ We propose at 5 plugins in this case:
These plugins have no requirements on which guest to host communication
plugin is used (you could, if you wanted, use 'direct serial' with
'cluster checkpoint', or 'multicast' with 'RHEV-H' for example).
'cluster cpg', or 'multicast' with 'RHEV-H' for example).
These plugins must also be able to discover where appropriate. For
example, the Checkpoint plugin can only be used if corosync/openais

2
man/fence_virt.8
View File

@ -9,7 +9,7 @@ fence_virt and fence_xvm are an I/O Fencing agents which can be used with virtua
Fence_virt and fence_xvm talk to fence_virtd, which supports multiple backend plugins, including:
- libvirt for single-node operation
- Cluster Checkpoints when using Linux-cluster release 3.0.0 or later
- Corosync CPG when using Linux-cluster release 3.0.0 or later
- libvirt-qpid for multi-node, non-cluster operation
For compatibility, fence_xvm may talk to fence_xvmd from linux-cluster

12
man/fence_virt.conf.5
View File

@ -236,21 +236,21 @@ Qpid service to connect to.
If set to 1, have fence_virtd use GSSAPI for authentication when communicating
with the Qpid broker. Default is 0 (off).
.SS checkpoint
.SS cpg
The checkpoint plugin uses CMAN, CPG, and OpenAIS checkpoints to track virtual
The cpg plugin uses corosync CPG and libvirt to track virtual
machines and route fencing requests to the appropriate computer.
.TP
.B uri
.
the URI to use when connecting to libvirt by the checkpoint plugin.
the URI to use when connecting to libvirt by the cpg plugin.
.TP
.B name_mode
.
The checkpoint plugin, in order to retain compatibility with fence_xvm,
stores virtual machines in a certain way in the OpenAIS checkpoints. The
The cpg plugin, in order to retain compatibility with fence_xvm,
stores virtual machines in a certain way. The
default was to use 'name' when using fence_xvm and fence_xvmd, and so this
is still the default. However, it is strongly recommended to use 'uuid'
instead of 'name' in all cluster environments involving more than one
@ -286,7 +286,7 @@ a key file.
fence_virtd {
listener = "multicast";
backend = "checkpoint";
backend = "cpg";
}
# this is the listeners section

2
man/fence_virtd.8
View File

@ -9,7 +9,7 @@ virtual machines.
Fence_virt and fence_xvm talk to fence_virtd, which supports multiple backend plugins, including:
- libvirt for single-node operation
- Cluster Checkpoints when using Linux-cluster release 3.0.0 or later
- Corosync CPG and libvirt when using Linux-cluster release 3.0.0 or later
- libvirt-qpid for multi-node, non-cluster operation
For compatibility, fence_xvm from linux-cluster release 2 may talk to fence_virtd.

25
server/Makefile.in
View File

@ -20,7 +20,6 @@ with_modules=@modules@
TARGETS=fence_virtd
MAIN_LIBS=-L../config -lsimpleconfig -ldl
AIS_LIBS=-L/usr/lib64/openais -lSaCkpt
COROSYNC_LIBS=-L/usr/lib64/corosync -lcpg
PACEMAKER_LIBS=-lcib -lcrmcommon -lpe_status -lplumb \
`pkg-config --libs ncurses` `pkg-config --libs glib-2.0`
@ -47,9 +46,9 @@ null_so_SOURCES = null.c
libvirt_qmf_so_SOURCES = uuid-test.c
libvirt_qmf_cxx_so_SOURCES = libvirt-qmf.cpp
pm_fence_so_SOURCES = pm-fence.c
cpg_so_SOURCES = cpg-virt.c cpg.c virt.c history.c uuid-test.c
multicast_so_SOURCES = mcast.c history.c
tcp_so_SOURCES = tcp.c history.c
checkpoint_so_SOURCES = virt.c vm_states.c history.c checkpoint.c cpg.c
serial_so_SOURCES = virt-serial.c virt-sockets.c serial.c history.c
@ -59,9 +58,9 @@ serial_so_SOURCES = virt-serial.c virt-sockets.c serial.c history.c
mod_null=@mod_null@
mod_libvirt=@mod_libvirt@
mod_checkpoint=@mod_checkpoint@
mod_libvirt_qmf=@mod_libvirt_qmf@
mod_pm_fence=@mod_pm_fence@
mod_cpg=@mod_cpg@
mod_multicast=@mod_multicast@
mod_serial=@mod_serial@
mod_tcp=@mod_tcp@
@ -77,9 +76,6 @@ MODULES=
ifneq ($(mod_libvirt),no)
MODULES+=libvirt.so
endif
ifneq ($(mod_checkpoint),no)
MODULES+=checkpoint.so
endif
ifneq ($(mod_libvirt_qmf),no)
MODULES+=libvirt-qmf.so
endif
@ -87,6 +83,9 @@ ifneq ($(mod_pm_fence),no)
INCLUDES+=$(PACEMAKER_INCLUDES)
MODULES+=pm-fence.so
endif
ifneq ($(mod_cpg),no)
MODULES+=cpg.so
endif
ifneq ($(mod_multicast),no)
MODULES+=multicast.so
endif
@ -110,10 +109,6 @@ ifneq ($(mod_libvirt),no)
fence_virtd_SOURCES+=${libvirt_so_SOURCES}
LIBS+=$(VIRT_LIBS) $(UUID_LIBS)
endif
ifneq ($(mod_checkpoint),no)
fence_virtd_SOURCES+=${checkpoint_so_SOURCES}
LIBS+=$(AIS_LIBS) $(COROSYNC_LIBS) $(CMAN_LIBS)
endif
ifneq ($(mod_libvirt_qmf),no)
fence_virtd_SOURCES+=${libvirt_qmf_so_SOURCES}
fence_virtd_cxx_SOURCES+=${libvirt_qmf_cxx_so_SOURCES}
@ -124,6 +119,9 @@ fence_virtd_SOURCES+=${pm_fence_so_SOURCES}
INCLUDES+=$(PACEMAKER_INCLUDES)
LIBS+=$(PACEMAKER_LIBS)
endif
ifneq ($(mod_cpg),no)
fence_virtd_SOURCES+=${cpg_so_SOURCES}
endif
ifneq ($(mod_multicast),no)
fence_virtd_SOURCES+=${multicast_so_SOURCES}
LIBS+=$(NSS_LIBS)
@ -166,13 +164,12 @@ libvirt-qmf.so: ${libvirt_qmf_so_SOURCES:.c=.o} ${libvirt_qmf_cxx_so_SOURCES:.cp
pm-fence.so: ${pm_fence_so_SOURCES:.c=.o}
$(CC) -o $@ $^ $(LIBS) -shared $(PACEMAKER_LIBS)
cpg.so: ${cpg_so_SOURCES:.c=.o}
$(CC) -o $@ $^ $(LIBS) -shared $(VIRT_LIBS) $(UUID_LIBS) $(XML_LIBS) $(COROSYNC_LIBS)
null.so: ${null_so_SOURCES:.c=.o}
$(CC) -o $@ $^ $(LIBS) -shared
checkpoint.so: ${checkpoint_so_SOURCES:.c=.o}
$(CC) -o $@ $^ $(LIBS) -shared $(AIS_LIBS) $(VIRT_LIBS) \
$(COROSYNC_LIBS) $(CMAN_LIBS)
serial.so: ${serial_so_SOURCES:.c=.o}
$(CC) -o $@ $^ $(LIBS) -shared $(VIRT_LIBS) $(UUID_LIBS) $(XML_LIBS)

824
server/checkpoint.c
View File

@ -1,824 +0,0 @@
/*
Copyright Red Hat, Inc. 2009
This program is free software; you can redistribute it and/or modify it
under the terms of the GNU General Public License as published by the
Free Software Foundation; either version 2, or (at your option) any
later version.
This program 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
General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program; see the file COPYING. If not, write to the
Free Software Foundation, Inc., 675 Mass Ave, Cambridge,
MA 02139, USA.
*/
/*
* Author: Lon Hohberger <lhh at redhat.com>
*/
#include <config.h>
#include <stdio.h>
#include <simpleconfig.h>
#include <static_map.h>
#include <sys/types.h>
#include <stdint.h>
#include <time.h>
#include <server_plugin.h>
#include <string.h>
#include <malloc.h>
#include <syslog.h>
#include <errno.h>
#include <unistd.h>
#include <libvirt/libvirt.h>
#include <pthread.h>
#ifdef HAVE_OPENAIS_CPG_H
#include <openais/cpg.h>
#else
#ifdef HAVE_COROSYNC_CPG_H
#include <corosync/cpg.h>
#endif
#endif
#include <libcman.h>
#include <debug.h>
#include "virt.h"
#include "xvm.h"
#include "checkpoint.h"
#define NAME "checkpoint"
#define VERSION "0.9"
#define MAGIC 0x1e017afe
struct check_info {
int magic;
int pad;
};
#define VALIDATE(arg) \
do {\
if (!arg || ((struct check_info *)arg)->magic != MAGIC) { \
errno = EINVAL;\
return -1; \
} \
} while(0)
static void *checkpoint_handle = NULL;
static virt_list_t *local_vms = NULL;
static char *uri = NULL;
static int use_uuid = 0;
static int
virt_list_update(virConnectPtr vp, virt_list_t **vl, int my_id)
{
virt_list_t *list = NULL;
if (*vl)
vl_free(*vl);
list = vl_get(vp, 1, my_id);
*vl = list;
if (!list)
return -1;
return 0;
}
static int
node_operational(uint32_t nodeid)
{
cman_handle_t ch;
cman_node_t node;
ch = cman_init(NULL);
if (!ch)
return -1;
memset(&node, 0, sizeof(node));
if (cman_get_node(ch, nodeid, &node) == 0) {
cman_finish(ch);
return !!node.cn_member;
}
cman_finish(ch);
return 0;
}
static int
get_domain_state_ckpt(void *hp, const char *domain, vm_state_t *state)
{
errno = EINVAL;
if (!hp || !domain || !state || !strlen((char *)domain))
return -1;
if (!strcmp(DOMAIN0NAME, (char *)domain))
return -1;
return ckpt_read(hp, domain, state, sizeof(*state));
}
static inline int
wait_domain(const char *vm_name, virConnectPtr vp, int timeout)
{
int tries = 0;
int response = 1;
int ret;
virDomainPtr vdp;
virDomainInfo vdi;
if (use_uuid) {
vdp = virDomainLookupByUUIDString(vp, (const char *)vm_name);
} else {
vdp = virDomainLookupByName(vp, vm_name);
}
if (!vdp)
return 0;
/* Check domain liveliness. If the domain is still here,
we return failure, and the client must then retry */
/* XXX On the xen 3.0.4 API, we will be able to guarantee
synchronous virDomainDestroy, so this check will not
be necessary */
do {
if (++tries > timeout)
break;
sleep(1);
if (use_uuid) {
vdp = virDomainLookupByUUIDString(vp,
(const char *)vm_name);
} else {
vdp = virDomainLookupByName(vp, vm_name);
}
if (!vdp) {
dbg_printf(2, "Domain no longer exists\n");
response = 0;
break;
}
memset(&vdi, 0, sizeof(vdi));
ret = virDomainGetInfo(vdp, &vdi);
virDomainFree(vdp);
if (ret < 0)
continue;
if (vdi.state == VIR_DOMAIN_SHUTOFF) {
dbg_printf(2, "Domain has been shut off\n");
response = 0;
break;
}
dbg_printf(4, "Domain still exists (state %d) "
"after %d seconds\n",
vdi.state, tries);
} while (1);
return response;
}
/*
Returns: 0 - operational
1 - dead or presumed so
2 - VM not local and I am not the right node to deal with it
3 - VM status unknown; cannot operate on it
*/
static int
cluster_virt_status(const char *vm_name, uint32_t *owner)
{
vm_state_t chk_state, temp_state;
virt_state_t *vs;
uint32_t me, high_id;
int ret = 0;
dbg_printf(80, "%s %s\n", __FUNCTION__, vm_name);
/* if we can't find the high ID, we can't do anything useful */
if (cpg_get_ids(&me, &high_id) != 0)
return 2;
if (use_uuid) {
vs = vl_find_uuid(local_vms, vm_name);
} else {
vs = vl_find_name(local_vms, vm_name);
}
if (!vs) {
ret = 2; /* not found locally */
temp_state.s_owner = 0;
temp_state.s_state = 0;
if (get_domain_state_ckpt(checkpoint_handle,
vm_name, &chk_state) < 0) {
if (me == high_id) {
dbg_printf(2, "High ID: Unknown VM\n");
ret = 3;
goto out;
}
} else if (me == chk_state.s_owner) {
/* <UVT> If domain has disappeared completely from libvirt (i.e., destroyed)
we'd end up with the checkpoing section containing its last state and last owner.
fence_virtd will freeze at the next status call, as no one will be willing to
return anything but 2. So we should delete corresponding section, but only if
we are high_id, because otherwise we don't know if the domain hasn't been started
on some other node. If checkpoint states us as an owner of the domain, but we
don't have it, we set s_state to a special value to let high_id know about
this situation. </UVT> */
dbg_printf(2, "I am an owner of unexisting domain, mangling field\n");
temp_state.s_owner = me;
temp_state.s_state = -1;
if (ckpt_write(checkpoint_handle, vm_name,
&temp_state, sizeof(vm_state_t)) < 0)
dbg_printf(2, "error storing in %s\n", __FUNCTION__);
}
if (me != high_id)
goto out;
if ((chk_state.s_state == -1) || (temp_state.s_state == -1)) {
dbg_printf(2, "I am high id and state field is mangled, removing section\n");
ckpt_erase (checkpoint_handle, vm_name);
ret = 1;
goto out;
}
if (node_operational(chk_state.s_owner)) {
*owner = chk_state.s_owner;
dbg_printf(2, "High ID: Owner is operational\n");
ret = 2;
} else {
dbg_printf(2, "High ID: Owner is dead; returning 'off'\n");
ret = 1;
}
} else if (vs->v_state.s_state == VIR_DOMAIN_SHUTOFF) {
ret = 1; /* local and off */
}
out:
dbg_printf(80, "%s %s %d\n", __FUNCTION__, vm_name, ret);
return ret;
}
static void
store_domains_by_name(void *hp, virt_list_t *vl)
{
int x;
if (!vl)
return;
for (x = 0; x < vl->vm_count; x++) {
if (!strcmp(DOMAIN0NAME, vl->vm_states[x].v_name))
continue;
dbg_printf(2, "Storing %s\n", vl->vm_states[x].v_name);
if (ckpt_write(hp, vl->vm_states[x].v_name,
&vl->vm_states[x].v_state,
sizeof(vm_state_t)) < 0)
dbg_printf(2, "error storing in %s\n", __FUNCTION__);
}
}
static void
store_domains_by_uuid(void *hp, virt_list_t *vl)
{
int x;
if (!vl)
return;
for (x = 0; x < vl->vm_count; x++) {
if (!strcmp(DOMAIN0UUID, vl->vm_states[x].v_uuid))
continue;
dbg_printf(2, "Storing %s\n", vl->vm_states[x].v_uuid);
if (ckpt_write(hp, vl->vm_states[x].v_uuid,
&vl->vm_states[x].v_state,
sizeof(vm_state_t)) < 0)
dbg_printf(2, "error storing in %s\n", __FUNCTION__);
}
}
static void
update_local_vms(void)
{
virConnectPtr vp = NULL;
uint32_t my_id = 0;
cpg_get_ids(&my_id, NULL);
vp = virConnectOpen(uri);
if (!vp) {
syslog(LOG_ERR, "Failed to connect to hypervisor\n");
}
virt_list_update(vp, &local_vms, my_id);
vl_print(local_vms);
if (use_uuid)
store_domains_by_uuid(checkpoint_handle, local_vms);
else
store_domains_by_name(checkpoint_handle, local_vms);
if (vp) virConnectClose(vp);
}
/* <UVT>
Functions do_off and do_reboot should return error only if fencing
was actualy unsuccessful, i.e., domain was running and is still
running after fencing attempt. If domain is not running after fencing
(did not exist before or couldn't be started after), 0 should be returned
</UVT> */
static int
do_off(const char *vm_name)
{
virConnectPtr vp;
virDomainPtr vdp;
virDomainInfo vdi;
int ret = -1;
dbg_printf(5, "%s %s\n", __FUNCTION__, vm_name);
vp = virConnectOpen(uri);
if (!vp)
return 1;
if (use_uuid) {
vdp = virDomainLookupByUUIDString(vp,
(const char *)vm_name);
} else {
vdp = virDomainLookupByName(vp, vm_name);
}
if (!vdp) {
dbg_printf(2, "Nothing to do - domain does not exist\n");
return 0;
}
if (((virDomainGetInfo(vdp, &vdi) == 0) &&
(vdi.state == VIR_DOMAIN_SHUTOFF))) {
dbg_printf(2, "Nothing to do - domain is off\n");
virDomainFree(vdp);
return 0;
}
syslog(LOG_NOTICE, "Destroying domain %s\n", vm_name);
dbg_printf(2, "[OFF] Calling virDomainDestroy\n");
ret = virDomainDestroy(vdp);
if (ret < 0) {
syslog(LOG_NOTICE, "Failed to destroy domain: %d\n", ret);
printf("virDomainDestroy() failed: %d\n", ret);
ret = 1;
goto out;
}
if (ret) {
syslog(LOG_NOTICE,
"Domain %s still exists; fencing failed\n",
vm_name);
printf("Domain %s still exists; fencing failed\n", vm_name);
ret = 1;
goto out;
}
ret = 0;
out:
virConnectClose(vp);
return ret;
}
static int
do_reboot(const char *vm_name)
{
virConnectPtr vp;
virDomainPtr vdp, nvdp;
virDomainInfo vdi;
char *domain_desc;
int ret;
//uuid_unparse(vm_uuid, uu_string);
dbg_printf(5, "%s %s\n", __FUNCTION__, vm_name);
vp = virConnectOpen(uri);
if (!vp)
return 1;
if (use_uuid) {
vdp = virDomainLookupByUUIDString(vp,
(const char *)vm_name);
} else {
vdp = virDomainLookupByName(vp, vm_name);
}
if (!vdp) {
dbg_printf(2, "[libvirt:REBOOT] Nothing to "
"do - domain does not exist\n");
return 0;
}
if (((virDomainGetInfo(vdp, &vdi) == 0) &&
(vdi.state == VIR_DOMAIN_SHUTOFF))) {
dbg_printf(2, "[libvirt:REBOOT] Nothing to "
"do - domain is off\n");
virDomainFree(vdp);
return 0;
}
syslog(LOG_NOTICE, "Rebooting domain %s\n", vm_name);
printf("Rebooting domain %s...\n", vm_name);
domain_desc = virDomainGetXMLDesc(vdp, 0);
if (!domain_desc) {
printf("Failed getting domain description from "
"libvirt\n");
}
dbg_printf(2, "[REBOOT] Calling virDomainDestroy(%p)\n", vdp);
ret = virDomainDestroy(vdp);
if (ret < 0) {
printf("virDomainDestroy() failed: %d/%d\n", ret, errno);
free(domain_desc);
virDomainFree(vdp);
ret = 1;
goto out;
}
ret = wait_domain(vm_name, vp, 15);
if (ret) {
syslog(LOG_NOTICE, "Domain %s still exists; fencing failed\n",
vm_name);
printf("Domain %s still exists; fencing failed\n", vm_name);
if (domain_desc)
free(domain_desc);
ret = 1;
goto out;
}
if (!domain_desc) {
ret = 0;
goto out;
}
/* 'on' is not a failure */
ret = 0;
dbg_printf(3, "[[ XML Domain Info ]]\n");
dbg_printf(3, "%s\n[[ XML END ]]\n", domain_desc);
dbg_printf(2, "Calling virDomainCreateLinux()...\n");
nvdp = virDomainCreateLinux(vp, domain_desc, 0);
if (nvdp == NULL) {
/* More recent versions of libvirt or perhaps the
* KVM back-end do not let you create a domain from
* XML if there is already a defined domain description
* with the same name that it knows about. You must
* then call virDomainCreate() */
dbg_printf(2, "Failed; Trying virDomainCreate()...\n");
if (virDomainCreate(vdp) < 0) {
syslog(LOG_NOTICE,
"Could not restart %s\n",
vm_name);
dbg_printf(1, "Failed to recreate guest"
" %s!\n", vm_name);
}
}
free(domain_desc);
out:
virConnectClose(vp);
return ret;
}
/*<UVT> This function must send reply from at least one node, otherwise
requesting fence_virtd would block forever in wait_cpt_reply </UVT> */
static void
do_real_work(void *data, size_t len, uint32_t nodeid, uint32_t seqno)
{
struct ckpt_fence_req *req = data;
struct ckpt_fence_req reply;
uint32_t owner;
int ret = 1;
memcpy(&reply, req, sizeof(reply));
update_local_vms();
switch(req->request) {
case FENCE_STATUS:
ret = cluster_virt_status(req->vm_name, &owner);
if (ret == 3) {
ret = RESP_OFF;
break;
}
if (ret == 2) {
return;
}
if (ret == 1) {
ret = RESP_OFF;
}
break;
case FENCE_OFF:
ret = cluster_virt_status(req->vm_name, &owner);
if (ret == 3) {
/* No record of this VM in the checkpoint. */
ret = 0;
break;
}
if (ret == 2) {
return;
}
if (ret == 1) {
ret = 0;
break;
}
/* Must be running locally to perform 'off' */
ret = do_off(req->vm_name);
break;
case FENCE_REBOOT:
ret = cluster_virt_status(req->vm_name, &owner);
if (ret == 3) {
ret = 0;
break;
}
if (ret == 2) {
return;
}
if (ret == 1) {
ret = 0;
break;
}
/* Must be running locally to perform 'reboot' */
ret = do_reboot(req->vm_name);
break;
}
reply.response = ret;
cpg_send_reply(&reply, sizeof(reply), nodeid, seqno);
}
static int
do_request(const char *vm_name, int request, uint32_t seqno)
{
struct ckpt_fence_req freq, *frp;
size_t retlen;
uint32_t seq;
int ret;
memset(&freq, 0, sizeof(freq));
snprintf(freq.vm_name, sizeof(freq.vm_name), vm_name);
freq.request = request;
freq.seqno = seqno;
if (cpg_send_req(&freq, sizeof(freq), &seq) != 0) {
printf("Failed to send\n");
return 1;
}
if (cpg_wait_reply((void *)&frp, &retlen, seq) != 0) {
printf("Failed to receive\n");
return 1;
}
ret = frp->response;
free(frp);
return ret;
}
static int
checkpoint_null(const char *vm_name, void *priv)
{
VALIDATE(priv);
printf("[CKPT] Null operation on %s\n", vm_name);
return 1;
}
static int
checkpoint_off(const char *vm_name, const char *src,
uint32_t seqno, void *priv)
{
VALIDATE(priv);
printf("[CKPT] OFF operation on %s seq %d\n", vm_name, seqno);
return do_request(vm_name, FENCE_OFF, seqno);
}
static int
checkpoint_on(const char *vm_name, const char *src,
uint32_t seqno, void *priv)
{
VALIDATE(priv);
printf("[CKPT] ON operation on %s seq %d\n", vm_name, seqno);
return 1;
}
static int
checkpoint_devstatus(void *priv)
{
printf("[CKPT] Device status\n");
VALIDATE(priv);
return 0;
}
static int
checkpoint_status(const char *vm_name, void *priv)
{
VALIDATE(priv);
printf("[CKPT] STATUS operation on %s\n", vm_name);
return do_request(vm_name, FENCE_STATUS, 0);
}
static int
checkpoint_reboot(const char *vm_name, const char *src,
uint32_t seqno, void *priv)
{
VALIDATE(priv);
printf("[CKPT] REBOOT operation on %s seq %d\n", vm_name, seqno);
return do_request(vm_name, FENCE_REBOOT, 0);
}
static int
checkpoint_hostlist(hostlist_callback callback, void *arg, void *priv)
{
VALIDATE(priv);
printf("[CKPT] HOSTLIST operation\n");
return 1;
}
static int
checkpoint_init(backend_context_t *c, config_object_t *config)
{
char value[1024];
struct check_info *info = NULL;
int x;
#ifdef _MODULE
if (sc_get(config, "fence_virtd/@debug", value, sizeof(value))==0)
dset(atoi(value));
#endif
if (sc_get(config, "backends/libvirt/@uri",
value, sizeof(value)) == 0) {
uri = strdup(value);
if (!uri) {
free(info);
return -1;
}
dbg_printf(1, "Using %s\n", uri);
}
if (sc_get(config, "backends/checkpoint/@uri",
value, sizeof(value)) == 0) {
if (uri)
free(uri);
uri = strdup(value);
if (!uri) {
free(info);
return -1;
}
dbg_printf(1, "Using %s\n", uri);
}
/* Naming scheme is no longer a top-level config option.
* However, we retain it here for configuration compatibility with
* versions 0.1.3 and previous.
*/
if (sc_get(config, "fence_virtd/@name_mode",
value, sizeof(value)-1) == 0) {
dbg_printf(1, "Got %s for name_mode\n", value);
if (!strcasecmp(value, "uuid")) {
use_uuid = 1;
} else if (!strcasecmp(value, "name")) {
use_uuid = 0;
} else {
dbg_printf(1, "Unsupported name_mode: %s\n", value);
}
}
if (sc_get(config, "backends/checkpoint/@name_mode",
value, sizeof(value)-1) == 0) {
dbg_printf(1, "Got %s for name_mode\n", value);
if (!strcasecmp(value, "uuid")) {
use_uuid = 1;
} else if (!strcasecmp(value, "name")) {
use_uuid = 0;
} else {
dbg_printf(1, "Unsupported name_mode: %s\n", value);
}
}
if (cpg_start(PACKAGE_NAME, do_real_work) < 0) {
return -1;
}
info = malloc(sizeof(*info));
if (!info)
return -1;
memset(info, 0, sizeof(*info));
info->magic = MAGIC;
x = 0;
while ((checkpoint_handle = ckpt_init(
"vm_states", 262144, 4096, 64, 10
)) == NULL) {
if (!x) {
dbg_printf(1, "Could not initialize "
"saCkPt; retrying...\n");
x = 1;
}
sleep(3);
}
if (x)
dbg_printf(1, "Checkpoint initialized\n");
update_local_vms();
*c = (void *)info;
return 0;
}
static int
checkpoint_shutdown(backend_context_t c)
{
struct check_info *info = (struct check_info *)c;
VALIDATE(info);
info->magic = 0;
free(info);
cpg_stop();
return 0;
}
static fence_callbacks_t checkpoint_callbacks = {
.null = checkpoint_null,
.off = checkpoint_off,
.on = checkpoint_on,
.reboot = checkpoint_reboot,
.status = checkpoint_status,
.devstatus = checkpoint_devstatus,
.hostlist = checkpoint_hostlist
};
static backend_plugin_t checkpoint_plugin = {
.name = NAME,
.version = VERSION,
.callbacks = &checkpoint_callbacks,
.init = checkpoint_init,
.cleanup = checkpoint_shutdown,
};
#ifdef _MODULE
double
BACKEND_VER_SYM(void)
{
return PLUGIN_VERSION_BACKEND;
}
const backend_plugin_t *
BACKEND_INFO_SYM(void)
{
return &checkpoint_plugin;
}
#else
static void __attribute__((constructor))
checkpoint_register_plugin(void)
{
plugin_reg_backend(&checkpoint_plugin);
}
#endif

21
server/config.c
View File

@ -129,36 +129,35 @@ backend_config_libvirt(config_object_t *config)
static int
backend_config_checkpoint(config_object_t *config)
backend_config_cpg(config_object_t *config)
{
char val[4096];
char inp[4096];
int done = 0;
printf("\n");
printf("The checkpoint backend module is designed for use in clusters\n"
"running corosync, openais, and CMAN. It utilizes the SAF \n"
"checkpoint API to store virtual machine states and CPG to \n"
printf("The CPG backend module is designed for use in clusters\n"
"running corosync and libvirt. It utilizes the CPG API to \n"
"route fencing requests, finally utilizing libvirt to perform\n"
"fencing actions.\n\n");
if (sc_get(config, "backends/checkpoint/@uri", val,
if (sc_get(config, "backends/cpg/@uri", val,
sizeof(val))) {
strncpy(val, DEFAULT_HYPERVISOR_URI, sizeof(val));
}
text_input("Libvirt URI", val, inp, sizeof(inp));
sc_set(config, "backends/checkpoint/@uri", inp);
sc_set(config, "backends/cpg/@uri", inp);
printf("\n");
printf("The name mode is how the checkpoint plugin stores and \n"
printf("The name mode is how the cpg plugin stores and \n"
"references virtual machines. Since virtual machine names\n"
"are not guaranteed to be unique cluster-wide, use of UUIDs\n"
"is strongly recommended. However, for compatibility with \n"
"fence_xvmd, the use of 'name' mode is also supported.\n\n");
if (sc_get(config, "backends/checkpoint/@name_mode", val,
if (sc_get(config, "backends/cpg/@name_mode", val,
sizeof(val))) {
strncpy(val, "uuid", sizeof(val));
}
@ -179,7 +178,7 @@ backend_config_checkpoint(config_object_t *config)
}
} while (!done);
sc_set(config, "backends/checkpoint/@name_mode", inp);
sc_set(config, "backends/cpg/@name_mode", inp);
return 0;
}
@ -527,8 +526,8 @@ backend_configure(config_object_t *config)
#if 0
if (!strcmp(inp, "libvirt")) {
backend_config_libvirt(config);
} else if (!strcmp(inp, "checkpoint")) {
backend_config_checkpoint(config);
} else if (!strcmp(inp, "cpg")) {
backend_config_cpg(config);
}
#endif

266
server/cpg.c
View File

@ -12,18 +12,14 @@
#include <sys/uio.h>
#include <list.h>
#include <pthread.h>
#ifdef HAVE_OPENAIS_CPG_H
#include <openais/cpg.h>
#else
#ifdef HAVE_COROSYNC_CPG_H
#include <corosync/cpg.h>
#endif
#endif
#include "checkpoint.h"
#define NODE_ID_NONE ((uint32_t)-1)
#include "debug.h"
#include "virt.h"
#include "cpg.h"
#define NODE_ID_NONE ((uint32_t) -1)
struct msg_queue_node {
list_head();
@ -38,6 +34,7 @@ struct msg_queue_node {
struct wire_msg {
#define TYPE_REQUEST 0
#define TYPE_REPLY 1
#define TYPE_STORE_VM 2
uint32_t type;
uint32_t seqno;
uint32_t target;
@ -45,95 +42,55 @@ struct wire_msg {
char data[0];
};
static uint32_t seqnum = 0, my_node_id = NODE_ID_NONE;
static uint32_t high_id_from_callback = NODE_ID_NONE;
static struct msg_queue_node *pending= NULL;
static uint32_t seqnum = 0;
static struct msg_queue_node *pending = NULL;
static cpg_handle_t cpg_handle;
static struct cpg_name gname;
static pthread_mutex_t cpg_mutex = PTHREAD_MUTEX_INITIALIZER;
static pthread_cond_t cpg_cond = PTHREAD_COND_INITIALIZER;
static pthread_t cpg_thread = 0;
static request_callback_fn req_callback_fn;
/* <UVT> function cpg_membership_get is (probably) buggy and returns correct
count only before cpg_mcast_joined, subsequent calls set count to 0 </UVT> */
#if 0
static pthread_mutex_t cpg_ids_mutex = PTHREAD_MUTEX_INITIALIZER;
static uint32_t my_node_id = NODE_ID_NONE;
static uint32_t high_id_from_callback = NODE_ID_NONE;
static request_callback_fn req_callback_fn;
static request_callback_fn store_callback_fn;
static confchange_callback_fn conf_leave_fn;
static confchange_callback_fn conf_join_fn;
int
cpg_get_ids(uint32_t *my_id, uint32_t *high_id)
{
/* This is segfaulting for some reason */
struct cpg_address cpg_nodes[CPG_MEMBERS_MAX];
uint32_t high = my_node_id;
int count = CPG_MEMBERS_MAX, x;
if (!my_id && !high_id)
return 0;
if (my_id)
*my_id = my_node_id;
if (!high_id)
return 0;
memset(&cpg_nodes, 0, sizeof(cpg_nodes));
if (cpg_membership_get(cpg_handle, &gname,
cpg_nodes, &count) != CPG_OK)
return -1;
for (x = 0; x < count; x++) {
if (cpg_nodes[x].nodeid > high) {
high = cpg_nodes[x].nodeid;
}
}
*high_id = high;
return 0;
}
#endif
int
cpg_get_ids(uint32_t *my_id, uint32_t *high_id)
{
if (!my_id && !high_id)
return 0;
pthread_mutex_lock(&cpg_ids_mutex);
if (my_id)
*my_id = my_node_id;
if (!high_id)
return 0;
*high_id = high_id_from_callback;
if (high_id)
*high_id = high_id_from_callback;
pthread_mutex_unlock(&cpg_ids_mutex);
return 0;
}
void
#ifdef HAVE_OPENAIS_CPG_H
cpg_deliver_func(cpg_handle_t h,
struct cpg_name *group_name,
uint32_t nodeid,
uint32_t pid,
void *msg,
int msglen)
#else
cpg_deliver_func(cpg_handle_t h,
const struct cpg_name *group_name,
uint32_t nodeid,
uint32_t pid,
void *msg,
size_t msglen)
#endif
{
struct msg_queue_node *n;
struct wire_msg *m = msg;
int x, found;
pthread_mutex_lock(&cpg_mutex);
if (m->type == TYPE_REPLY) {
/* Reply to a request we sent */
found = 0;
@ -165,10 +122,7 @@ cpg_deliver_func(cpg_handle_t h,
list_remove(&pending, n);
list_insert(&pending, n);
#if 0
printf("Seqnum %d replied; removing from list",
n->seqno);
#endif
dbg_printf(2, "Seqnum %d replied; removing from list\n", n->seqno);
pthread_cond_broadcast(&cpg_cond);
goto out_unlock;
@ -179,6 +133,10 @@ cpg_deliver_func(cpg_handle_t h,
req_callback_fn(&m->data, msglen - sizeof(*m),
nodeid, m->seqno);
}
if (m->type == TYPE_STORE_VM) {
store_callback_fn(&m->data, msglen - sizeof(*m),
nodeid, m->seqno);
}
return;
@ -188,32 +146,31 @@ out_unlock:
void
#ifdef HAVE_OPENAIS_CPG_H
cpg_config_change(cpg_handle_t h,
struct cpg_name *group_name,
struct cpg_address *members, int memberlen,
struct cpg_address *left, int leftlen,
struct cpg_address *join, int joinlen)
#else
cpg_config_change(cpg_handle_t h,
const struct cpg_name *group_name,
const struct cpg_address *members, size_t memberlen,
const struct cpg_address *left, size_t leftlen,
const struct cpg_address *join, size_t joinlen)
#endif
{
int x;
int high = my_node_id;
int high;
pthread_mutex_lock(&cpg_ids_mutex);
high = my_node_id;
for (x = 0; x < memberlen; x++) {
if (members[x].nodeid > high) {
if (members[x].nodeid > high)
high = members[x].nodeid;
}
}
high_id_from_callback = high;
pthread_mutex_unlock(&cpg_ids_mutex);
return;
if (joinlen > 0)
conf_join_fn(join, joinlen);
if (leftlen > 0)
conf_leave_fn(left, leftlen);
}
@ -235,9 +192,12 @@ cpg_send_req(void *data, size_t len, uint32_t *seqno)
n = malloc(sizeof(*n));
if (!n)
return -1;
m = malloc(msgsz);
if (!m)
if (!m) {
free(n);
return -1;
}
/* only incremented on send */
n->state = STATE_CLEAR;
@ -260,15 +220,58 @@ cpg_send_req(void *data, size_t len, uint32_t *seqno)
ret = cpg_mcast_joined(cpg_handle, CPG_TYPE_AGREED, &iov, 1);
free(m);
if (ret == CPG_OK)
if (ret == CS_OK)
return 0;
return -1;
}
int
cpg_send_reply(void *data, size_t len, uint32_t nodeid,
uint32_t seqno)
cpg_send_vm_state(virt_state_t *vs)
{
struct iovec iov;
struct msg_queue_node *n;
struct wire_msg *m;
size_t msgsz = sizeof(*m) + sizeof(*vs);
int ret;
n = calloc(1, (sizeof(*n)));
if (!n)
return -1;
m = calloc(1, msgsz);
if (!m) {
free(n);
return -1;
}
n->state = STATE_MESSAGE;
n->msg = NULL;
n->msglen = 0;
pthread_mutex_lock(&cpg_mutex);
list_insert(&pending, n);
pthread_mutex_unlock(&cpg_mutex);
m->type = TYPE_STORE_VM;
m->target = NODE_ID_NONE;
memcpy(&m->data, vs, sizeof(*vs));
iov.iov_base = m;
iov.iov_len = msgsz;
ret = cpg_mcast_joined(cpg_handle, CPG_TYPE_AGREED, &iov, 1);
free(m);
if (ret == CS_OK)
return 0;
return -1;
}
int
cpg_send_reply(void *data, size_t len, uint32_t nodeid, uint32_t seqno)
{
struct iovec iov;
struct wire_msg *m;
@ -290,8 +293,9 @@ cpg_send_reply(void *data, size_t len, uint32_t nodeid,
ret = cpg_mcast_joined(cpg_handle, CPG_TYPE_AGREED, &iov, 1);
free(m);
if (ret == CPG_OK)
if (ret == CS_OK)
return 0;
return -1;
}
@ -313,11 +317,12 @@ cpg_wait_reply(void **data, size_t *len, uint32_t seqno)
if (n->state != STATE_MESSAGE)
continue;
found = 1;
break;
goto out;
}
pthread_mutex_unlock(&cpg_mutex);
}
out:
list_remove(&pending, n);
pthread_mutex_unlock(&cpg_mutex);
@ -332,54 +337,60 @@ cpg_wait_reply(void **data, size_t *len, uint32_t seqno)
static void *
cpg_dispatch_thread(void *arg)
{
cpg_dispatch(cpg_handle, CPG_DISPATCH_BLOCKING);
cpg_dispatch(cpg_handle, CS_DISPATCH_BLOCKING);
return NULL;
}
int
cpg_start(const char *name, request_callback_fn func)
cpg_start( const char *name,
request_callback_fn req_cb_fn,
request_callback_fn store_cb_fn,
confchange_callback_fn join_fn,
confchange_callback_fn leave_fn)
{
cpg_handle_t h;
int ret;
errno = EINVAL;
if (!name)
return -1;
gname.length = snprintf(gname.value,
sizeof(gname.value), name);
if (gname.length >= sizeof(gname.value)) {
ret = snprintf(gname.value, sizeof(gname.value), name);
if (ret <= 0)
return -1;
if (ret >= sizeof(gname.value)) {
errno = ENAMETOOLONG;
return -1;
}
if (gname.length <= 0)
return -1;
gname.length = ret;
memset(&h, 0, sizeof(h));
if (cpg_initialize(&h, &my_callbacks) != CPG_OK) {
if (cpg_initialize(&h, &my_callbacks) != CS_OK) {
perror("cpg_initialize");
return -1;
}
if (cpg_join(h, &gname) != CPG_OK) {
if (cpg_join(h, &gname) != CS_OK) {
perror("cpg_join");
return -1;
}
cpg_local_get(h, &my_node_id);
dbg_printf(2, "My CPG nodeid is %d\n", my_node_id);
pthread_mutex_lock(&cpg_mutex);
cpg_local_get(h, &my_node_id);
pthread_create(&cpg_thread, NULL, cpg_dispatch_thread, NULL);
memcpy(&cpg_handle, &h, sizeof(h));
req_callback_fn = func;
req_callback_fn = req_cb_fn;
store_callback_fn = store_cb_fn;
conf_join_fn = join_fn;
conf_leave_fn = leave_fn;
pthread_mutex_unlock(&cpg_mutex);
@ -392,59 +403,8 @@ cpg_stop(void)
{
pthread_cancel(cpg_thread);
pthread_join(cpg_thread, NULL);
cpg_leave(cpg_handle, &gname);
cpg_finalize(cpg_handle);
return 0;
}
#ifdef STANDALONE
int please_quit = 0;
void
go_away(int sig)
{
please_quit = 1;
}
void
request_callback(void *data, size_t len, uint32_t nodeid, uint32_t seqno)
{
char *msg = data;
printf("msg = %s\n", msg);
cpg_send_reply("fail.", 7, nodeid, seqno);
}
int
main(int argc, char **argv)
{
uint32_t seqno = 0;
int fd;
char *data;
size_t len;
signal(SIGINT, go_away);
if (cpg_start("lhh1", request_callback) < 0) {
perror("cpg_start");
return 1;
}
cpg_send_req("hi", 2, &seqno);
cpg_wait_reply(&data, &len, seqno);
printf("%s\n", data);
printf("going bye\n");
cpg_stop();
return 0;
}
#endif

16
server/checkpoint.h → server/cpg.h
View File

@ -1,7 +1,7 @@
#ifndef _CHECKPOINT_H
#define _CHECKPOINT_H
#ifndef __FENCE_VIRTD_CPG_H
#define __FENCE_VIRTD_CPG_H
struct ckpt_fence_req {
struct cpg_fence_req {
char vm_name[128];
int request;
uint32_t seqno;
@ -10,12 +10,20 @@ struct ckpt_fence_req {
typedef void (*request_callback_fn)(void *data, size_t len, uint32_t nodeid,
uint32_t seqno);
typedef void (*confchange_callback_fn)(const struct cpg_address *m, size_t len);
int cpg_start( const char *name,
request_callback_fn func,
request_callback_fn store_func,
confchange_callback_fn join,
confchange_callback_fn leave);
int cpg_get_ids(uint32_t *me, uint32_t *high);
int cpg_start(const char *name, request_callback_fn func);
int cpg_stop(void);
int cpg_send_req(void *data, size_t len, uint32_t *seqno);
int cpg_wait_reply(void **data, size_t *len, uint32_t seqno);
int cpg_send_reply(void *data, size_t len, uint32_t nodeid, uint32_t seqno);
int cpg_send_vm_state(virt_state_t *vs);
#endif

101
server/virt.c
View File

@ -68,17 +68,20 @@ virt_list_t *vl_get(virConnectPtr *vp, int vp_count, int my_id)
for (i = 0 ; i < vp_count ; i++) {
int x;
virDomainPtr *dom_list;
virt_list_t *new_vl;
int ret = virConnectListAllDomains(vp[i], &dom_list, 0);
if (ret <= 0)
continue;
d_count += ret;
vl = realloc(vl, sizeof(uint32_t) + sizeof(virt_state_t) * d_count);
if (!vl) {
new_vl = realloc(vl, sizeof(uint32_t) + sizeof(virt_state_t) * d_count);
if (!new_vl) {
_free_dom_list(dom_list, ret);
free(vl);
return NULL;
}
vl = new_vl;
vl->vm_count = d_count;
/* Ok, we have the domain IDs - let's get their names and states */
@ -123,6 +126,100 @@ virt_list_t *vl_get(virConnectPtr *vp, int vp_count, int my_id)
return vl;
}
int
vl_add(virt_list_t **vl, virt_state_t *vm) {
virt_list_t *new_vl;
size_t oldlen;
size_t newlen;
if (!vl)
return -1;
if (!*vl) {
*vl = malloc(sizeof(uint32_t) + sizeof(virt_state_t));
if (!*vl)
return -1;
(*vl)->vm_count = 1;
memcpy(&(*vl)->vm_states[0], vm, sizeof(virt_state_t));
return 0;
}
oldlen = sizeof(uint32_t) + sizeof(virt_state_t) * (*vl)->vm_count;
newlen = oldlen + sizeof(virt_state_t);
new_vl = malloc(newlen);
if (!new_vl)
return -1;
memcpy(new_vl, *vl, oldlen);
memcpy(&new_vl->vm_states[(*vl)->vm_count], vm, sizeof(virt_state_t));
new_vl->vm_count++;
free(*vl);
*vl = new_vl;
return 0;
}
int vl_remove_by_owner(virt_list_t **vl, uint32_t owner) {
int i;
int removed = 0;
virt_list_t *new_vl;
if (!vl || !*vl)
return 0;
for (i = 0 ; i < (*vl)->vm_count ; i++) {
if ((*vl)->vm_states[i].v_state.s_owner == owner) {
dbg_printf(2, "Removing %s\n", (*vl)->vm_states[i].v_name);
memset(&(*vl)->vm_states[i].v_state, 0,
sizeof((*vl)->vm_states[i].v_state));
(*vl)->vm_states[i].v_name[0] = 0xff;
(*vl)->vm_states[i].v_uuid[0] = 0xff;
removed++;
}
}
if (!removed)
return 0;
qsort(&(*vl)->vm_states[0], (*vl)->vm_count, sizeof((*vl)->vm_states[0]),
_compare_virt);
(*vl)->vm_count -= removed;
new_vl = realloc(*vl, sizeof(uint32_t) + (sizeof(virt_state_t) * ((*vl)->vm_count)));
if (new_vl)
*vl = new_vl;
return removed;
}
int
vl_update(virt_list_t **vl, virt_state_t *vm) {
virt_state_t *v = NULL;
if (!vl)
return -1;
if (!*vl)
return vl_add(vl, vm);
if (strlen(vm->v_uuid) > 0)
v = vl_find_uuid(*vl, vm->v_uuid);
if (v == NULL && strlen(vm->v_name) > 0)
v = vl_find_name(*vl, vm->v_name);
if (v == NULL) {
dbg_printf(2, "Adding new entry for VM %s\n", vm->v_name);
vl_add(vl, vm);
} else {
dbg_printf(2, "Updating entry for VM %s\n", vm->v_name);
memcpy(&v->v_state, &vm->v_state, sizeof(v->v_state));
}
return 0;
}
void
vl_print(virt_list_t *vl)

28
server/virt.h
View File

@ -26,22 +26,6 @@
#include "xvm.h"
/*
Owner 0 = no owner.
checkpoint "xen-vm-states" {
section "vm-name0" {
owner_nodeid;
vm_state;
}
section "vm-name1" {
owner_nodeid;
vm_state;
}
...
}
*/
typedef struct {
uint32_t s_owner;
int32_t s_state;
@ -66,19 +50,13 @@ void vl_print(virt_list_t *vl);
void vl_free(virt_list_t *old);
virt_state_t *vl_find_uuid(virt_list_t *vl, const char *name);
virt_state_t *vl_find_name(virt_list_t *vl, const char *name);
int vl_add(virt_list_t **vl, virt_state_t *vm);
int vl_update(virt_list_t **vl, virt_state_t *vm);
int vl_remove_by_owner(virt_list_t **vl, uint32_t owner);
int vm_off(virConnectPtr *vp, int vp_count, const char *vm_name);
int vm_on(virConnectPtr *vp, int vp_count, const char *vm_name);
int vm_status(virConnectPtr *vp, int vp_count, const char *vm_name);
int vm_reboot(virConnectPtr *vp, int vp_count, const char *vm_name);
typedef void ckpt_handle;
int ckpt_read(void *hp, const char *secid, void *buf, size_t maxlen);
int ckpt_finish(void *hp);
int ckpt_write(void *hp, const char *secid, void *buf, size_t maxlen);
int ckpt_erase(void *hp, const char *secid);
void *ckpt_init(const char *ckpt_name, int maxlen, int maxsec, int maxseclen,
int timeout);
#endif

458
server/vm_states.c
View File

@ -1,458 +0,0 @@
/*
Copyright Red Hat, Inc. 2006
This program is free software; you can redistribute it and/or modify it
under the terms of the GNU General Public License as published by the
Free Software Foundation; either version 2, or (at your option) any
later version.
This program 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
General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program; see the file COPYING. If not, write to the
Free Software Foundation, Inc., 675 Mass Ave, Cambridge,
MA 02139, USA.
*/
/** @file
* Distributed VM states using saCkpt interface
*/
#include <string.h>
#include <errno.h>
#include <sys/types.h>
#include <sys/resource.h>
#include <sys/wait.h>
#include <stdlib.h>
#include <sys/time.h>
#include <pthread.h>
#include <openais/saAis.h>
#include <openais/saCkpt.h>
#include <unistd.h>
#include <stdio.h>
#include <assert.h>
#include "xvm.h"
typedef struct {
uint32_t ck_ready;
int ck_timeout;
SaCkptCheckpointHandleT ck_checkpoint;
SaCkptHandleT ck_handle;
char *ck_name;
} ckpt_handle;
#define READY_MAGIC 0x13fd237c
#define VALIDATE(h) \
do { \
if (!h || h->ck_ready != READY_MAGIC) { \
errno = EINVAL; \
return -1; \
} \
} while(0)
int ais_to_posix(SaAisErrorT err);
int
ais_to_posix(SaAisErrorT err)
{
switch (err) {
case SA_AIS_OK:
return 0;
case SA_AIS_ERR_LIBRARY:
return ELIBBAD;
case SA_AIS_ERR_VERSION:
return EPROTONOSUPPORT; //XXX
case SA_AIS_ERR_INIT:
return EFAULT; //XXX
case SA_AIS_ERR_TIMEOUT:
return ETIMEDOUT;
case SA_AIS_ERR_TRY_AGAIN:
return EAGAIN;
case SA_AIS_ERR_INVALID_PARAM:
return EINVAL;
case SA_AIS_ERR_NO_MEMORY:
return ENOMEM;
case SA_AIS_ERR_BAD_HANDLE:
return EBADF;
case SA_AIS_ERR_BUSY:
return EBUSY;
case SA_AIS_ERR_ACCESS:
return EACCES;
case SA_AIS_ERR_NOT_EXIST:
return ENOENT;
case SA_AIS_ERR_NAME_TOO_LONG:
return ENAMETOOLONG;
case SA_AIS_ERR_EXIST:
return EEXIST;
case SA_AIS_ERR_NO_SPACE:
return ENOSPC;
case SA_AIS_ERR_INTERRUPT:
return EINTR;
case SA_AIS_ERR_NAME_NOT_FOUND:
return ENOENT;
case SA_AIS_ERR_NO_RESOURCES:
return ENOMEM; //XXX
case SA_AIS_ERR_NOT_SUPPORTED:
return ENOSYS;
case SA_AIS_ERR_BAD_OPERATION:
return EINVAL; //XXX
case SA_AIS_ERR_FAILED_OPERATION:
return EIO; //XXX
case SA_AIS_ERR_MESSAGE_ERROR:
return EIO; // XXX
case SA_AIS_ERR_QUEUE_FULL:
return ENOBUFS;
case SA_AIS_ERR_QUEUE_NOT_AVAILABLE:
return ENOENT;
case SA_AIS_ERR_BAD_FLAGS:
return EINVAL;
case SA_AIS_ERR_TOO_BIG:
return E2BIG;
case SA_AIS_ERR_NO_SECTIONS:
return ENOENT; // XXX
/*case SA_AIS_ERR_SECURITY:
return EPERM;*/
default:
return EINVAL; /* XXX */
}
return -1;
}
static int
ckpt_open(ckpt_handle *h, const char *ckpt_name, int maxsize,
int maxsec, int maxsecsize, int timeout)
{
SaCkptCheckpointCreationAttributesT attrs;
SaCkptCheckpointOpenFlagsT flags;
SaNameT cpname;
#if 0
SaCkptCheckpointDescriptorT status;
#endif
SaAisErrorT err = SA_AIS_OK;
VALIDATE(h);
flags = SA_CKPT_CHECKPOINT_READ |
SA_CKPT_CHECKPOINT_WRITE;
snprintf((char *)cpname.value, SA_MAX_NAME_LENGTH-1,
"%s", ckpt_name);
cpname.length = strlen(ckpt_name);
h->ck_timeout = timeout;
err = saCkptCheckpointOpen(h->ck_handle,
&cpname,
NULL,
flags,
timeout,
&h->ck_checkpoint);
if (err == SA_AIS_OK) {
#if 0
saCkptCheckpointStatusGet(h->ck_handle,
&status);
printf("Checkpoint Size = %d bytes\n", (int)
status.checkpointCreationAttributes.checkpointSize);
printf("Flags = ");
if (status.checkpointCreationAttributes.creationFlags &
SA_CKPT_WR_ALL_REPLICAS) {
printf("%s ", "SA_CKPT_WR_ALL_REPLICAS");
}
if (status.checkpointCreationAttributes.creationFlags &
SA_CKPT_WR_ACTIVE_REPLICA) {
printf("%s ", "SA_CKPT_WR_ACTIVE_REPLICA");
}
if (status.checkpointCreationAttributes.creationFlags &
SA_CKPT_WR_ACTIVE_REPLICA_WEAK) {
printf("%s ", "SA_CKPT_WR_ACTIVE_REPLICA_WEAK");
}
if (status.checkpointCreationAttributes.creationFlags &
SA_CKPT_CHECKPOINT_COLLOCATED) {
printf("%s ", "SA_CKPT_CHECKPOINT_COLLOCATED");
}
printf("\nMax sections = %d\n",
(int)status.checkpointCreationAttributes.maxSections);
printf("Max section size = %d\n",
(int)status.checkpointCreationAttributes.maxSectionSize);
printf("Max section ID size = %d\n",
(int)status.checkpointCreationAttributes.maxSectionIdSize);
printf("Section count = %d\n", status.numberOfSections);
printf("\n");
#endif
goto good;
}
attrs.creationFlags = SA_CKPT_WR_ALL_REPLICAS;
attrs.checkpointSize = (SaSizeT)maxsize;
attrs.retentionDuration = SA_TIME_ONE_MINUTE;
attrs.maxSections = maxsec;
attrs.maxSectionSize = (SaSizeT)maxsecsize;
attrs.maxSectionIdSize = (SaSizeT)MAX_DOMAINNAME_LENGTH;
flags = SA_CKPT_CHECKPOINT_READ |
SA_CKPT_CHECKPOINT_WRITE |
SA_CKPT_CHECKPOINT_CREATE;
err = saCkptCheckpointOpen(h->ck_handle,
&cpname,
&attrs,
flags,
timeout,
&h->ck_checkpoint);
if (err == SA_AIS_OK)
goto good;
/* No checkpoint */
errno = ais_to_posix(err);
return (errno == 0 ? 0 : -1);
good:
printf("Opened ckpt %s\n", ckpt_name);
h->ck_name = strdup(ckpt_name);
errno = ais_to_posix(err);
return (errno == 0 ? 0 : -1);
}
void *
ckpt_init(char *ckpt_name, int maxlen, int maxsec,
int maxseclen, int timeout)
{
ckpt_handle *h;
SaAisErrorT err;
SaVersionT ver;
if (!ckpt_name || !strlen(ckpt_name)) {
errno = EINVAL;
return NULL;
}
h = malloc(sizeof(*h));
if (!h)
return NULL;
memset(h, 0, sizeof(*h));
ver.releaseCode = 'B';
ver.majorVersion = 1;
ver.minorVersion = 1;
err = saCkptInitialize(&h->ck_handle, NULL, &ver);
if (err != SA_AIS_OK) {
free(h);
return NULL;
} else {
h->ck_ready = READY_MAGIC;
}
if (ckpt_open(h, ckpt_name, maxlen, maxsec, maxseclen,
timeout) < 0) {
saCkptCheckpointClose(h->ck_checkpoint);
if (h->ck_name)
free(h->ck_name);
free(h);
return NULL;
}
return (void *)h;
}
int
ckpt_write(void *hp, const char *secid, void *buf, size_t maxlen)
{
ckpt_handle *h = (ckpt_handle *)hp;
SaCkptIOVectorElementT iov = {SA_CKPT_DEFAULT_SECTION_ID,
NULL, 0, 0, 0};
SaAisErrorT err;
SaCkptSectionCreationAttributesT attrs;
VALIDATE(h);
/* Set section ID here */
iov.sectionId.id = (uint8_t *)secid;
iov.sectionId.idLen = strlen(secid);
iov.dataBuffer = buf;
iov.dataSize = (SaSizeT)maxlen;
iov.dataOffset = 0;
iov.readSize = 0;
err = saCkptCheckpointWrite(h->ck_checkpoint, &iov, 1, NULL);
if (err == SA_AIS_ERR_NOT_EXIST) {
attrs.sectionId = &iov.sectionId;
attrs.expirationTime = SA_TIME_END;
err = saCkptSectionCreate(h->ck_checkpoint, &attrs,
buf, maxlen);
}
if (err == SA_AIS_OK)
saCkptCheckpointSynchronize(h->ck_checkpoint,
h->ck_timeout);
errno = ais_to_posix(err);
if (errno)
return -1;
return maxlen; /* XXX */
}
int
ckpt_read(void *hp, const char *secid, void *buf, size_t maxlen)
{
ckpt_handle *h = (ckpt_handle *)hp;
SaCkptIOVectorElementT iov = {SA_CKPT_DEFAULT_SECTION_ID,
NULL, 0, 0, 0};
SaAisErrorT err;
VALIDATE(h);
//printf("reading ckpt %s\n", keyid);
iov.sectionId.id = (uint8_t *)secid;
iov.sectionId.idLen = strlen(secid);
iov.dataBuffer = buf;
iov.dataSize = (SaSizeT)maxlen;
iov.dataOffset = 0;
iov.readSize = 0;
err = saCkptCheckpointRead(h->ck_checkpoint, &iov, 1, NULL);
errno = ais_to_posix(err);
if (errno)
return -1;
return iov.readSize; /* XXX */
}
int
ckpt_erase(void *hp, const char *secid)
{
ckpt_handle *h = (ckpt_handle *)hp;
SaAisErrorT err;
SaCkptSectionIdT sectionId;
VALIDATE(h);
sectionId.id = (uint8_t *)secid;
sectionId.idLen = strlen(secid);
err = saCkptSectionDelete(h->ck_checkpoint, &sectionId);
if (err == SA_AIS_OK)
saCkptCheckpointSynchronize(h->ck_checkpoint,
h->ck_timeout);
errno = ais_to_posix(err);
if (errno)
return -1;
return 0;
}
int
ckpt_finish(void *hp)
{
ckpt_handle *h = (ckpt_handle *)hp;
int ret = 0;
SaAisErrorT err;
saCkptCheckpointClose(h->ck_checkpoint);
err = saCkptFinalize(h->ck_handle);
if (err != SA_AIS_OK)
ret = -1;
else
h->ck_ready = 0;
if (h->ck_name)
free(h->ck_name);
if (ret != 0)
errno = ais_to_posix(err);
return ret;
}
#ifdef STANDALONE
void
usage(int ret)
{
printf("usage: ckpt [-c ckpt_name] <-r key|-w key -d data>\n");
exit(ret);
}
int
main(int argc, char **argv)
{
char *ckptname = "ckpt_test";
char *sec = "default";
char *val;
void *h;
char buf[64];
int ret;
int op = 0;
while((ret = getopt(argc, argv, "c:w:r:d:j?")) != EOF) {
switch(ret) {
case 'c':
ckptname = optarg;
break;
case 'w':
op = 'w';
sec = optarg;
break;
case 'r':
op = 'r';
sec = optarg;
break;
case 'd':
val = optarg;
break;
case '?':
case 'h':
usage(0);
default:
usage(1);
}
}
if (!op) {
usage(1);
}
if (!sec) {
usage(1);
}
h = ckpt_init(ckptname, 262144, 4096, 64, 10);
if (!h) {
perror("ckpt_init");
return -1;
}
if (op == 'w') {
if (ckpt_write(h, sec, val, strlen(val)+1) < 0) {
perror("ckpt_write");
return 1;
}
} else if (op == 'r') {
ret = ckpt_read(h, sec, buf, sizeof(buf));
if (ret < 0) {
perror("ckpt_read");
return 1;
}
printf("%d bytes\nDATA for '%s':\n%s\n", ret, sec,
buf);
}
ckpt_finish(h);
return 0;
}
#endif