[Devel] [PATCH 8/8] vzctl: checkpoint/restoer of upstream CTs

[Kir Kolyshkin]

On 05/15/2013 09:10 AM, Andrey Vagin wrote:
> It can be done by any third party tools.
> sripts/ct-cpt and script/ct_rst are executed to dump and restore CT
> This patch adds scripts for CRIU.
>
> Signed-off-by: Andrey Vagin <avagin at openvz.org>
> ---
>   scripts/ct-cpt     | 19 ++++++++++++
>   scripts/ct-rst     | 17 ++++++++++
>   src/lib/hooks_ct.c | 91 +++++++++++++++++++++++++++++++++++++++++++++++++++++-
>   3 files changed, 126 insertions(+), 1 deletion(-)
>   create mode 100755 scripts/ct-cpt
>   create mode 100755 scripts/ct-rst
>
> diff --git a/scripts/ct-cpt b/scripts/ct-cpt
> new file mode 100755
> index 0000000..87c09dc
> --- /dev/null
> +++ b/scripts/ct-cpt
> @@ -0,0 +1,19 @@
> +#!/bin/sh
> +mkdir $CT_CR_DIR &&

You'd better
1 describe what environment variables you expect here
2 check for their existence, refusing to run otherwise

> +crtools dump	--file-locks		\

It's criu :)

> +		--tcp-established	\
> +		--evasive-devices	\
> +		--link-remap		\
> +		--root $CT_ROOT		\

I'd keep the old name, ie VE_ROOT, since this is what all the other 
scripts use all over.

> +		-t $CT_PID		\
> +		-D $CT_CR_DIR		\
> +		-o dump.log		\
> +		-vvvv
> +if [ $? -ne 0 ]; then
> +	[ -d $CT_CR_DIR.fail ] && rm -rf $CT_CR_DIR.fail
> +	mv -f $CT_CR_DIR $CT_CR_DIR.fail
> +	echo All dump files and logs were saved in $CT_CR_DIR.fail
> +	exit 1
> +else
> +	echo Checkpointing finished successfully
> +fi
> diff --git a/scripts/ct-rst b/scripts/ct-rst
> new file mode 100755
> index 0000000..8953ec6
> --- /dev/null
> +++ b/scripts/ct-rst
> @@ -0,0 +1,17 @@
> +#!/bin/sh
> +crtools restore	--file-locks		\

criu

> +		--tcp-established	\
> +		--evasive-devices	\
> +		--link-remap		\
> +		--root $CT_ROOT		\
> +		--restore-detached	\
> +		-D $CT_CR_DIR		\
> +		-o restore.log		\
> +		-vvvv			\
> +		--pidfile $CT_STATE
> +if [ $? -eq 0 ]; then
> +	rm -rf $CT_CR_DIR
> +else
> +	echo The restore log was saved in $CT_CR_DIR/restore.log
> +	exit 1
> +fi
> diff --git a/src/lib/hooks_ct.c b/src/lib/hooks_ct.c
> index 3c82823..97c8695 100644
> --- a/src/lib/hooks_ct.c
> +++ b/src/lib/hooks_ct.c
> @@ -17,6 +17,7 @@
>   #include "logger.h"
>   #include "script.h"
>   #include "cgroup.h"
> +#include "cpt.h"
>   
>   #define NETNS_RUN_DIR "/var/run/netns"
>   
> @@ -307,7 +308,10 @@ int ct_env_create(struct arg_start *arg)
>   		return VZ_RESOURCE_ERROR;
>   	}
>   
> -	ret = ct_env_create_real(arg);
> +	if (arg->fn)
> +		ret = arg->fn(arg->h, arg->veid, arg->res, arg->wait_p, arg->old_wait_p, arg->err_p, arg->data);
> +	else
> +		ret = ct_env_create_real(arg);
>   
>   	snprintf(procpath, STR_SIZE, "/proc/%d/ns/net", ret);
>   	ret = symlink(procpath, ctpath);
> @@ -597,6 +601,89 @@ static int ct_setcontext(envid_t veid)
>   	return 0;
>   }
>   
> +static int ct_chkpnt(vps_handler *h, envid_t veid, const vps_res *res,
> +		int cmd, cpt_param *param)
> +{
> +	const char *dumpfile = NULL;
> +	char pidstr[STR_SIZE], buf[STR_SIZE];
> +	int ret, fd;
> +	char *arg[2];
> +	char *env[4];
> +
> +	ret = VZ_CHKPNT_ERROR;
> +
> +	get_dump_file(veid, param->dumpdir, buf, sizeof(buf));
> +	dumpfile = strdup(buf);
> +
> +	arg[0] = SCRIPTDIR "/ct-cpt";
> +	arg[1] = NULL;
> +
> +	get_state_file(veid, res->env.statedir, pidstr, sizeof(pidstr));
> +	fd = open(pidstr, O_RDONLY);
> +	if (fd == -1) {
> +		logger(-1, 0, "Unable to open the pid file");

second logger() argument should be errno not 0 here (and in every place 
where errno makes sense).

> +		return VZ_CHKPNT_ERROR;
> +	}
> +
> +	ret = read(fd, pidstr, sizeof(pidstr) - 1);
> +	if (ret == -1) {
> +		logger(-1, 0, "Unable to read pid");

ditto

> +		return VZ_CHKPNT_ERROR;
> +	}
> +	pidstr[ret] = '\0';
> +
> +	snprintf(buf, sizeof(buf), "CT_ROOT=%s", res->fs.root);
VE_ROOT
> +	env[0] = strdup(buf);
> +	snprintf(buf, sizeof(buf), "CT_PID=%s", pidstr);
> +	env[1] = strdup(buf);
> +	snprintf(buf, sizeof(buf), "CT_CR_DIR=%s", dumpfile);
> +	env[2] = strdup(buf);
> +	env[3] = NULL;
> +
> +	if (run_script(arg[0], arg, env, 0))
> +		return ret;
> +
> +	return 0;
> +}
> +
> +static int ct_env_restore(vps_handler *h, envid_t veid, vps_res *res,
> +			  int wait_p, int old_wait_p, int err_p, void *data)
> +{
> +	char *env[4];
> +	char *argv[2];
> +	char buf[STR_SIZE];
> +	const char *dumpfile = NULL;
> +	const char *statefile = NULL;
> +	cpt_param *param = data;
> +
> +	get_dump_file(veid, param->dumpdir, buf, sizeof(buf));
> +	dumpfile = strdup(buf);
> +
> +	get_state_file(veid, res->env.statedir, buf, sizeof(buf));
> +	statefile = strdup(buf);
> +
> +	argv[0] = SCRIPTDIR "/ct-rst";
> +	argv[1] = NULL;
> +
> +	snprintf(buf, sizeof(buf), "CT_ROOT=%s", res->fs.root);
> +	env[0] = strdup(buf);
> +	snprintf(buf, sizeof(buf), "CT_CR_DIR=%s", dumpfile);
> +	env[1] = strdup(buf);
> +	snprintf(buf, sizeof(buf), "CT_STATE=%s", statefile);

Bad choice for a name. We have VE_STATE for other scripts, which can be 
"running", "starting" etc.

> +	env[2] = strdup(buf);
> +	env[3] = NULL;
> +	if (run_script(argv[0], argv, env, 0))
> +		return VZ_RESTORE_ERROR;
> +
> +	return 0;
> +}
> +
> +static int ct_restore(vps_handler *h, envid_t veid, vps_param *vps_p, int cmd,
> +	cpt_param *param, skipFlags skip)
> +{
> +	return vps_start_custom(h, veid, vps_p, SKIP_CONFIGURE | skip, NULL, ct_env_restore, param);
> +}
> +
>   int ct_do_open(vps_handler *h)
>   {
>   	int ret;
> @@ -630,6 +717,8 @@ int ct_do_open(vps_handler *h)
>   	h->enter = ct_enter;
>   	h->destroy = ct_destroy;
>   	h->env_create = ct_env_create;
> +	h->env_chkpnt = ct_chkpnt;
> +	h->env_restore = ct_restore;
>   	h->setlimits = ct_setlimits;
>   	h->setcpus = ct_setcpus;
>   	h->setcontext = ct_setcontext;