[CRIU] Re: [PATCH 5/5] util: move network socket to free FD during
restore
Pavel Emelyanov
xemul at parallels.com
Tue Aug 21 12:39:24 EDT 2012
On 08/21/2012 07:20 PM, Adrian Reber wrote:
> The socket listening on the network has to be moved during the
> restore so that it does not collide with FDs which are trying
> to be restored.
>
> Signed-off-by: Adrian Reber <adrian at lisas.de>
> ---
> image.c | 7 +++++++
> include/crtools.h | 1 +
> util.c | 9 +++++++++
> 3 files changed, 17 insertions(+), 0 deletions(-)
>
> diff --git a/image.c b/image.c
> index 969017a..0d84c5c 100644
> --- a/image.c
> +++ b/image.c
> @@ -366,3 +366,10 @@ int reopen_listen_socket()
> return open_listen_socket(network_host, network_port);
> return 0;
> }
> +
> +void update_listen_socket(int fd)
> +{
> + if (sockfd == -1)
> + return;
> + sockfd = fd;
> +}
> diff --git a/include/crtools.h b/include/crtools.h
> index c092802..f056327 100644
> --- a/include/crtools.h
> +++ b/include/crtools.h
> @@ -135,6 +135,7 @@ extern void close_image_dir(void);
> extern int open_listen_socket(char *host, int port);
> extern void close_listen_socket();
> extern int reopen_listen_socket();
> +extern void update_listen_socket(int fd);
>
> int open_image(int type, unsigned long flags, ...);
> #define open_image_ro(type, ...) open_image(type, O_RDONLY, ##__VA_ARGS__)
> diff --git a/util.c b/util.c
> index 4969e8d..91795d1 100644
> --- a/util.c
> +++ b/util.c
> @@ -83,6 +83,15 @@ int reopen_fd_as_safe(char *file, int line, int new_fd, int old_fd, bool allow_r
> if (old_fd != new_fd) {
>
> if (!allow_reuse_fd) {
> + /* see if we are running a network migration */
> + tmp = open_listen_socket(NULL, 0);
> + if (tmp != -1) {
> + int tmp2;
> + /* yes, network socket is open; let's move it */
> + tmp2 = fcntl(tmp, F_DUPFD, new_fd);
> + update_listen_socket(tmp2);
> + close(tmp);
You should just place a call to move_img_fd to proper place to prevent
it from being overwritten by the restore fd.
> + }
> if (fcntl(new_fd, F_GETFD) != -1 || errno != EBADF) {
> if (new_fd < 3) {
> /*
More information about the CRIU
mailing list