|
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] Re: [Xen-devel] [PATCH] libxenlight: fix suspend\resume
On Fri, 27 Nov 2009, Stefano Stabellini wrote:
> Hi all,
> this patch fixes the current suspend\resume implementation in
> libxenlight and creates the correspondent commands in xl.
>
> The patch applies after Tomasz' console patch.
>
> Signed-off-by: Stefano Stabellini <stefano.stabellini@xxxxxxxxxxxxx>
>
resending because the previous version had a bug if the device model
state was missing.
Signed-off-by: Stefano Stabellini <stefano.stabellini@xxxxxxxxxxxxx>
---
diff -r c34a58c843a6 tools/libxl/libxl.c
--- a/tools/libxl/libxl.c Fri Nov 27 16:24:45 2009 +0000
+++ b/tools/libxl/libxl.c Fri Nov 27 16:37:36 2009 +0000
@@ -217,29 +217,36 @@
}
int libxl_domain_restore(struct libxl_ctx *ctx, libxl_domain_build_info *info,
- uint32_t domid, int fd)
+ uint32_t domid, int fd, libxl_domain_build_state
*state,
+ libxl_device_model_info *dm_info)
{
- libxl_domain_build_state state;
char **vments = NULL, **localents = NULL;
- memset(&state, '\0', sizeof(state));
-
- build_pre(ctx, domid, info, &state);
- restore_common(ctx, domid, info, &state, fd);
+ build_pre(ctx, domid, info, state);
+ restore_common(ctx, domid, info, state, fd);
if (info->hvm) {
- vments = libxl_calloc(ctx, 4, sizeof(char *));
+ vments = libxl_calloc(ctx, 5, sizeof(char *));
vments[0] = "rtc/timeoffset";
vments[1] = (info->u.hvm.timeoffset) ? info->u.hvm.timeoffset : "";
+ vments[2] = "image/ostype";
+ vments[3] = "hvm";
} else {
- localents = libxl_calloc(ctx, 4 * 2, sizeof(char *));
- localents[0] = "serial/0/limit";
- localents[1] = libxl_sprintf(ctx, "%d", 65536);
- localents[2] = "console/port";
- localents[3] = libxl_sprintf(ctx, "%d", state.console_port);
- localents[4] = "console/ring-ref";
- localents[5] = libxl_sprintf(ctx, "%ld", state.console_mfn);
+ vments = libxl_calloc(ctx, 9, sizeof(char *));
+ vments[0] = "image/ostype";
+ vments[1] = "linux";
+ vments[2] = "image/kernel";
+ vments[3] = (char*) info->kernel;
+ vments[4] = "image/ramdisk";
+ vments[5] = (char*) info->u.pv.ramdisk;
+ vments[6] = "image/cmdline";
+ vments[7] = (char*) info->u.pv.cmdline;
}
- build_post(ctx, domid, info, &state, vments, localents);
+ build_post(ctx, domid, info, state, vments, localents);
+ if (info->hvm)
+ asprintf(&(dm_info->saved_state), "/var/lib/xen/qemu-save.%d", domid);
+ else
+ dm_info->saved_state = NULL;
+
return 0;
}
@@ -299,17 +306,37 @@
return info;
}
+static int libxl_save_device_model(struct libxl_ctx *ctx, uint32_t domid, int
fd)
+{
+ int fd2, c;
+ char buf[1024];
+ char *filename = libxl_sprintf(ctx, "/var/lib/xen/qemu-save.%d", domid);
+
+ XL_LOG(ctx, XL_LOG_DEBUG, "Saving device model state to %s", filename);
+ libxl_xs_write(ctx, XBT_NULL, libxl_sprintf(ctx,
"/local/domain/0/device-model/%d/command", domid), "save", strlen("save"));
+ libxl_wait_for_device_model(ctx, domid, "paused", NULL, NULL);
+
+ write(fd, QEMU_SIGNATURE, strlen(QEMU_SIGNATURE));
+ fd2 = open(filename, O_RDONLY);
+ while ((c = read(fd2, buf, sizeof(buf))) != 0) {
+ write(fd, buf, c);
+ }
+ close(fd2);
+ unlink(filename);
+ return 0;
+}
+
int libxl_domain_suspend(struct libxl_ctx *ctx, libxl_domain_suspend_info
*info,
uint32_t domid, int fd)
{
- int hvm = 1;
- int live = 0;
- int debug = 0;
- char savesig[] = "XenSavedDomain\n";
+ int hvm = is_hvm(ctx, domid);
+ int live = info != NULL && info->flags & XL_SUSPEND_LIVE;
+ int debug = info != NULL && info->flags & XL_SUSPEND_LIVE;
- write(fd, savesig, strlen(savesig));
core_suspend(ctx, domid, fd, hvm, live, debug);
+ if (hvm)
+ libxl_save_device_model(ctx, domid, fd);
return 0;
}
@@ -322,7 +349,19 @@
int libxl_domain_unpause(struct libxl_ctx *ctx, uint32_t domid)
{
+ char path[50];
+ char *state;
+
+ if (is_hvm(ctx, domid)) {
+ snprintf(path, sizeof(path), "/local/domain/0/device-model/%d/state",
domid);
+ state = libxl_xs_read(ctx, XBT_NULL, path);
+ if (state != NULL && !strcmp(state, "paused")) {
+ libxl_xs_write(ctx, XBT_NULL, libxl_sprintf(ctx,
"/local/domain/0/device-model/%d/command", domid), "continue",
strlen("continue"));
+ libxl_wait_for_device_model(ctx, domid, "running", NULL, NULL);
+ }
+ }
xc_domain_unpause(ctx->xch, domid);
+
return 0;
}
@@ -581,6 +620,10 @@
vifs[i].devid, vifs[i].ifname, vifs[i].bridge));
}
}
+ }
+ if (info->saved_state) {
+ flexarray_set(dm_args, num++, "-loadvm");
+ flexarray_set(dm_args, num++, info->saved_state);
}
for (i = 0; info->extra && info->extra[i] != NULL; i++)
flexarray_set(dm_args, num++, info->extra[i]);
diff -r c34a58c843a6 tools/libxl/libxl.h
--- a/tools/libxl/libxl.h Fri Nov 27 16:24:45 2009 +0000
+++ b/tools/libxl/libxl.h Fri Nov 27 16:37:36 2009 +0000
@@ -94,6 +94,8 @@
} libxl_domain_build_state;
typedef struct {
+#define XL_SUSPEND_DEBUG 1
+#define XL_SUSPEND_LIVE 2
int flags;
int (*suspend_callback)(void *, int);
} libxl_domain_suspend_info;
@@ -107,6 +109,7 @@
int domid;
char *dom_name;
char *device_model;
+ char *saved_state;
libxl_qemu_machine_type type;
int videoram; /* size of the videoram in MB */
bool stdvga; /* stdvga enabled or disabled */
@@ -254,7 +257,8 @@
int libxl_domain_make(struct libxl_ctx *ctx, libxl_domain_create_info *info,
uint32_t *domid);
int libxl_domain_build(struct libxl_ctx *ctx, libxl_domain_build_info *info,
uint32_t domid, /* out */ libxl_domain_build_state *state);
int libxl_domain_restore(struct libxl_ctx *ctx, libxl_domain_build_info *info,
- uint32_t domid, int fd);
+ uint32_t domid, int fd, libxl_domain_build_state
*state,
+ libxl_device_model_info *dm_info);
int libxl_domain_suspend(struct libxl_ctx *ctx, libxl_domain_suspend_info
*info,
uint32_t domid, int fd);
int libxl_domain_shutdown(struct libxl_ctx *ctx, uint32_t domid, int req);
diff -r c34a58c843a6 tools/libxl/libxl_dom.c
--- a/tools/libxl/libxl_dom.c Fri Nov 27 16:24:45 2009 +0000
+++ b/tools/libxl/libxl_dom.c Fri Nov 27 16:37:36 2009 +0000
@@ -163,71 +163,36 @@
state->store_port, &state->store_mfn,
state->console_port, &state->console_mfn,
info->hvm, info->u.hvm.pae, 0);
+#if defined(__i386__) || defined(__x86_64__)
+ xc_cpuid_apply_policy(ctx->xch, domid);
+#endif
return 0;
}
-/* the following code is extremely ugly and racy without forking.
- we intend to fix the re-entrancy of the underlying code instead of forking
*/
-static struct libxl_ctx *global_suspend_ctx = NULL;
-static struct suspendinfo {
- int xch;
+struct suspendinfo {
+ struct libxl_ctx *ctx;
int xce; /* event channel handle */
int suspend_eventchn;
int domid;
int hvm;
unsigned int flags;
-} si;
+};
-void core_suspend_switch_qemu_logdirty(int domid, unsigned int enable)
+static void core_suspend_switch_qemu_logdirty(int domid, unsigned int enable)
{
- struct xs_handle *xs;
- char *path, *ret_path, *cmd_path, *ret_str, *cmd_str, **watch;
- unsigned int len;
- struct timeval tv;
- fd_set fdset;
- struct libxl_ctx *ctx = global_suspend_ctx;
+ struct xs_handle *xsh;
+ char path[64];
- xs = xs_daemon_open();
- if (!xs)
- return;
- path = libxl_sprintf(ctx, "/local/domain/0/device-model/%i/logdirty",
domid);
- if (!path)
- return;
- ret_path = libxl_sprintf(ctx, "%s/ret", path);
- if (!ret_path)
- return;
- cmd_path = libxl_sprintf(ctx, "%s/cmd", path);
- if (!ret_path)
- return;
+ snprintf(path, sizeof(path),
"/local/domain/0/device-model/%u/logdirty/cmd", domid);
- /* Watch for qemu's return value */
- if (!xs_watch(xs, ret_path, "qemu-logdirty-ret"))
- return;
+ xsh = xs_daemon_open();
- cmd_str = (enable == 0) ? "disable" : "enable";
+ if (enable)
+ xs_write(xsh, XBT_NULL, path, "enable", strlen("enable"));
+ else
+ xs_write(xsh, XBT_NULL, path, "disable", strlen("disable"));
- /* Tell qemu that we want it to start logging dirty page to Xen */
- if (!xs_write(xs, XBT_NULL, cmd_path, cmd_str, strlen(cmd_str)))
- return;
-
- /* Wait a while for qemu to signal that it has service logdirty command */
-read_again:
- tv.tv_sec = 5;
- tv.tv_usec = 0;
- FD_ZERO(&fdset);
- FD_SET(xs_fileno(xs), &fdset);
-
- if ((select(xs_fileno(xs) + 1, &fdset, NULL, NULL, &tv)) != 1)
- return;
-
- watch = xs_read_watch(xs, &len);
- free(watch);
-
- ret_str = xs_read(xs, XBT_NULL, ret_path, &len);
- if (ret_str == NULL || strcmp(ret_str, cmd_str))
- /* Watch fired but value is not yet right */
- goto read_again;
- free(ret_str);
+ xs_daemon_close(xsh);
}
static int core_suspend_callback(void *data)
@@ -235,46 +200,78 @@
struct suspendinfo *si = data;
unsigned long s_state = 0;
int ret;
+ char *path, *state = "suspend";
+ int watchdog = 60;
if (si->hvm)
- xc_get_hvm_param(si->xch, si->domid, HVM_PARAM_ACPI_S_STATE, &s_state);
+ xc_get_hvm_param(si->ctx->xch, si->domid, HVM_PARAM_ACPI_S_STATE,
&s_state);
if ((s_state == 0) && (si->suspend_eventchn >= 0)) {
- ret = xc_evtchn_notify(si->xch, si->suspend_eventchn);
+ ret = xc_evtchn_notify(si->xce, si->suspend_eventchn);
if (ret < 0) {
+ XL_LOG(si->ctx, XL_LOG_ERROR, "xc_evtchn_notify failed ret=%d",
ret);
return 0;
}
- ret = xc_await_suspend(si->xch, si->suspend_eventchn);
+ ret = xc_await_suspend(si->xce, si->suspend_eventchn);
if (ret < 0) {
+ XL_LOG(si->ctx, XL_LOG_ERROR, "xc_await_suspend failed ret=%d",
ret);
return 0;
}
return 1;
}
- /* need to shutdown (to suspend) the domain here */
- return 0;
+ path = libxl_sprintf(si->ctx, "%s/control/shutdown",
libxl_xs_get_dompath(si->ctx, si->domid));
+ libxl_xs_write(si->ctx, XBT_NULL, path, "suspend", strlen("suspend"));
+ if (si->hvm) {
+ unsigned long hvm_pvdrv, hvm_s_state;
+ xc_get_hvm_param(si->ctx->xch, si->domid, HVM_PARAM_CALLBACK_IRQ,
&hvm_pvdrv);
+ xc_get_hvm_param(si->ctx->xch, si->domid, HVM_PARAM_ACPI_S_STATE,
&hvm_s_state);
+ if (!hvm_pvdrv || hvm_s_state) {
+ XL_LOG(si->ctx, XL_LOG_DEBUG, "Calling xc_domain_shutdown on the
domain");
+ xc_domain_shutdown(si->ctx->xch, si->domid, SHUTDOWN_suspend);
+ }
+ }
+ XL_LOG(si->ctx, XL_LOG_DEBUG, "wait for the guest to suspend");
+ while (!strcmp(state, "suspend") && watchdog > 0) {
+ int nb_domain, i;
+ xc_dominfo_t *list = NULL;
+ usleep(100000);
+ list = libxl_domain_infolist(si->ctx, &nb_domain);
+ for (i = 0; i < nb_domain; i++) {
+ if (si->domid == list[i].domid) {
+ if (list[i].shutdown != 0 && list[i].shutdown_reason ==
SHUTDOWN_suspend) {
+ free(list);
+ return 1;
+ }
+ }
+ }
+ free(list);
+ state = libxl_xs_read(si->ctx, XBT_NULL, path);
+ watchdog--;
+ }
+ if (!strcmp(state, "suspend")) {
+ XL_LOG(si->ctx, XL_LOG_ERROR, "guest didn't suspend in time");
+ libxl_xs_write(si->ctx, XBT_NULL, path, "", 1);
+ }
+ return 1;
}
-static struct save_callbacks callbacks;
int core_suspend(struct libxl_ctx *ctx, uint32_t domid, int fd,
int hvm, int live, int debug)
{
int flags;
int port;
+ struct save_callbacks callbacks;
+ struct suspendinfo si;
flags = (live) ? XCFLAGS_LIVE : 0
- | (debug) ? XCFLAGS_DEBUG : 0;
-
- /* crappy global lock until we make everything clean */
- while (global_suspend_ctx) {
- sleep(1);
- }
- global_suspend_ctx = ctx;
+ | (debug) ? XCFLAGS_DEBUG : 0
+ | (hvm) ? XCFLAGS_HVM : 0;
si.domid = domid;
si.flags = flags;
si.hvm = hvm;
- si.suspend_eventchn = si.xce = -1;
- si.xch = ctx->xch;
+ si.ctx = ctx;
+ si.suspend_eventchn = -1;
si.xce = xc_evtchn_open();
if (si.xce < 0)
@@ -284,28 +281,28 @@
port = xs_suspend_evtchn_port(si.domid);
if (port < 0) {
+ XL_LOG(ctx, XL_LOG_WARNING, "Failed to get the suspend evtchn
port");
} else {
- si.suspend_eventchn = xc_suspend_evtchn_init(si.xch, si.xce,
si.domid, port);
+ si.suspend_eventchn = xc_suspend_evtchn_init(si.ctx->xch, si.xce,
si.domid, port);
- if (si.suspend_eventchn < 0) {
- }
+ if (si.suspend_eventchn < 0)
+ XL_LOG(ctx, XL_LOG_WARNING, "Suspend event channel
initialization failed");
}
}
+ memset(&callbacks, 0, sizeof(callbacks));
callbacks.suspend = core_suspend_callback;
- callbacks.postcopy = NULL;
- callbacks.checkpoint = NULL;
callbacks.data = &si;
xc_domain_save(ctx->xch, fd, domid, 0, 0, flags,
&callbacks, hvm,
- core_suspend_switch_qemu_logdirty);
+ &core_suspend_switch_qemu_logdirty);
if (si.suspend_eventchn > 0)
xc_suspend_evtchn_release(si.xce, si.suspend_eventchn);
if (si.xce > 0)
xc_evtchn_close(si.xce);
- global_suspend_ctx = NULL;
return 0;
}
+
diff -r c34a58c843a6 tools/libxl/libxl_internal.h
--- a/tools/libxl/libxl_internal.h Fri Nov 27 16:24:45 2009 +0000
+++ b/tools/libxl/libxl_internal.h Fri Nov 27 16:37:36 2009 +0000
@@ -30,6 +30,7 @@
#define LIBXL_DESTROY_TIMEOUT 10
#define LIBXL_XENCONSOLE_LIMIT 1048576
#define LIBXL_XENCONSOLE_PROTOCOL "vt100"
+#define QEMU_SIGNATURE "QemuDeviceModelRecord"
#define ARRAY_SIZE(a) (sizeof(a) / sizeof(a[0]))
diff -r c34a58c843a6 tools/libxl/xl.c
--- a/tools/libxl/xl.c Fri Nov 27 16:24:45 2009 +0000
+++ b/tools/libxl/xl.c Fri Nov 27 16:37:36 2009 +0000
@@ -31,6 +31,7 @@
#include <sys/select.h>
#include <arpa/inet.h>
#include <xenctrl.h>
+
#include "libxl.h"
#include "libxl_utils.h"
@@ -577,7 +578,7 @@
} \
})
-static void create_domain(int debug, const char *filename)
+static void create_domain(int debug, const char *config_file, const char
*restore_file, int paused)
{
struct libxl_ctx ctx;
uint32_t domid;
@@ -595,9 +596,10 @@
int i, fd;
int need_daemon = 1;
libxl_device_model_starting *dm_starting = 0;
+ memset(&dm_info, 0x00, sizeof(dm_info));
- printf("Parsing config file %s\n", filename);
- parse_config_file(filename, &info1, &info2, &disks, &num_disks, &vifs,
&num_vifs, &pcidevs, &num_pcidevs, &vfbs, &num_vfbs, &vkbs, &num_vkbs,
&dm_info);
+ printf("Parsing config file %s\n", config_file);
+ parse_config_file(config_file, &info1, &info2, &disks, &num_disks, &vifs,
&num_vifs, &pcidevs, &num_pcidevs, &vfbs, &num_vfbs, &vkbs, &num_vkbs,
&dm_info);
if (debug)
printf_info(&info1, &info2, disks, num_disks, vifs, num_vifs, pcidevs,
num_pcidevs, vfbs, num_vfbs, vkbs, num_vkbs, &dm_info);
@@ -607,7 +609,20 @@
libxl_ctx_init(&ctx);
libxl_ctx_set_log(&ctx, log_callback, NULL);
libxl_domain_make(&ctx, &info1, &domid);
- libxl_domain_build(&ctx, &info2, domid, &state);
+
+ if (!restore_file || !need_daemon) {
+ if (dm_info.saved_state) {
+ free(dm_info.saved_state);
+ dm_info.saved_state = NULL;
+ }
+ libxl_domain_build(&ctx, &info2, domid, &state);
+ } else {
+ int restore_fd;
+
+ restore_fd = open(restore_file, O_RDONLY);
+ libxl_domain_restore(&ctx, &info2, domid, restore_fd, &state,
&dm_info);
+ close(restore_fd);
+ }
for (i = 0; i < num_disks; i++) {
disk_info_domid_fixup(disks + i, domid);
@@ -642,7 +657,8 @@
for (i = 0; i < num_pcidevs; i++)
libxl_device_pci_add(&ctx, domid, &pcidevs[i]);
- libxl_domain_unpause(&ctx, domid);
+ if (!paused)
+ libxl_domain_unpause(&ctx, domid);
if (need_daemon) {
char *fullname, *name;
@@ -714,6 +730,8 @@
printf(" pause pause execution of a
domain\n\n");
printf(" unpause unpause a paused domain\n\n");
printf(" console attach to domain's
console\n\n");
+ printf(" save save a domain state to restore
later\n\n");
+ printf(" restore restore a domain from a saved
state\n\n");
} else if(!strcmp(command, "create")) {
printf("Usage: xl create <ConfigFile> [options] [vars]\n\n");
printf("Create a domain based on <ConfigFile>.\n\n");
@@ -738,6 +756,18 @@
} else if(!strcmp(command, "unpause")) {
printf("Usage: xl unpause <Domain>\n\n");
printf("Unpause a paused domain.\n\n");
+ } else if(!strcmp(command, "save")) {
+ printf("Usage: xl save [options] <Domain> <CheckpointFile>\n\n");
+ printf("Save a domain state to restore later.\n\n");
+ printf("Options:\n\n");
+ printf("-h Print this help.\n");
+ printf("-c Leave domain running after creating the
snapshot.\n");
+ } else if(!strcmp(command, "restore")) {
+ printf("Usage: xl restore [options] <ConfigFile>
<CheckpointFile>\n\n");
+ printf("Restore a domain from a saved state.\n\n");
+ printf("Options:\n\n");
+ printf("-h Print this help.\n");
+ printf("-p Do not unpause domain after restoring
it.\n");
} else if(!strcmp(command, "destroy")) {
printf("Usage: xl destroy <Domain>\n\n");
printf("Terminate a domain immediately.\n\n");
@@ -1017,6 +1047,101 @@
free(info);
}
+int save_domain(char *p, char *filename, int checkpoint)
+{
+ struct libxl_ctx ctx;
+ uint32_t domid;
+ int fd;
+
+ libxl_ctx_init(&ctx);
+ libxl_ctx_set_log(&ctx, log_callback, NULL);
+
+ if (libxl_param_to_domid(&ctx, p, &domid) < 0) {
+ fprintf(stderr, "%s is an invalid domain identifier\n", p);
+ exit(2);
+ }
+ fd = open(filename, O_WRONLY|O_CREAT|O_TRUNC, 0644);
+ if (fd < 0) {
+ fprintf(stderr, "Failed to open temp file %s for writing\n", filename);
+ exit(2);
+ }
+ libxl_domain_suspend(&ctx, NULL, domid, fd);
+ close(fd);
+
+ if (checkpoint)
+ libxl_domain_unpause(&ctx, domid);
+ else
+ libxl_domain_destroy(&ctx, domid, 0);
+
+ exit(0);
+}
+
+int main_restore(int argc, char **argv)
+{
+ char *checkpoint_file = NULL;
+ char *config_file = NULL;
+ int paused = 0, debug = 0;
+ int opt;
+
+ while ((opt = getopt(argc, argv, "hpd")) != -1) {
+ switch (opt) {
+ case 'p':
+ paused = 1;
+ break;
+ case 'd':
+ debug = 1;
+ break;
+ case 'h':
+ help("restore");
+ exit(0);
+ default:
+ fprintf(stderr, "option not supported\n");
+ break;
+ }
+ }
+
+ if (optind >= argc - 1) {
+ help("restore");
+ exit(2);
+ }
+
+ config_file = argv[optind];
+ checkpoint_file = argv[optind + 1];
+ create_domain(debug, config_file, checkpoint_file, paused);
+ exit(0);
+}
+
+int main_save(int argc, char **argv)
+{
+ char *filename = NULL, *p = NULL;
+ int checkpoint = 0;
+ int opt;
+
+ while ((opt = getopt(argc, argv, "hc")) != -1) {
+ switch (opt) {
+ case 'c':
+ checkpoint = 1;
+ break;
+ case 'h':
+ help("save");
+ exit(0);
+ default:
+ fprintf(stderr, "option not supported\n");
+ break;
+ }
+ }
+
+ if (optind >= argc - 1) {
+ help("save");
+ exit(2);
+ }
+
+ p = argv[optind];
+ filename = argv[optind + 1];
+ save_domain(p, filename, checkpoint);
+ exit(0);
+}
+
int main_pause(int argc, char **argv)
{
int opt;
@@ -1142,7 +1267,7 @@
}
filename = argv[optind];
- create_domain(debug, filename);
+ create_domain(debug, filename, NULL, 0);
exit(0);
}
@@ -1171,6 +1296,10 @@
main_unpause(argc - 1, argv + 1);
} else if (!strcmp(argv[1], "console")) {
main_console(argc - 1, argv + 1);
+ } else if (!strcmp(argv[1], "save")) {
+ main_save(argc - 1, argv + 1);
+ } else if (!strcmp(argv[1], "restore")) {
+ main_restore(argc - 1, argv + 1);
} else if (!strcmp(argv[1], "help")) {
if (argc > 2)
help(argv[2]);
_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel
|
 |
Lists.xenproject.org is hosted with RackSpace, monitoring our |