eventpoll: Add checkpoint/restore v2

v2:
 - Move everything into eventpoll.[ch]
 - Use rst_file_params
Signed-off-by: Cyrill Gorcunov <gorcunov@openvz.org>
Signed-off-by: Pavel Emelyanov <xemul@parallels.com>

Makefile

@@ -52,6 +52,7 @@ OBJS		+= ipc_ns.o
 OBJS		+= netfilter.o
 OBJS		+= shmem.o
 OBJS		+= eventfd.o
+OBJS		+= eventpoll.o
 
 DEPS		:= $(patsubst %.o,%.d,$(OBJS))
 

cr-dump.c

@@ -38,6 +38,7 @@
 #include "shmem.h"
 #include "sk-inet.h"
 #include "eventfd.h"
+#include "eventpoll.h"
 
 #ifndef CONFIG_X86_64
 # error No x86-32 support yet
@@ -325,6 +326,9 @@ static int do_dump_gen_file(const struct fd_parms *p, int lfd,
 		case FDINFO_EVENTFD:
 			ret = dump_one_eventfd(lfd, e.id, p);
 			break;
+		case FDINFO_EVENTPOLL:
+			ret = dump_one_eventpoll(lfd, e.id, p);
+			break;
 		default:
 			ret = dump_one_reg_file(lfd, e.id, p);
 			break;
@@ -464,6 +468,13 @@ static int dump_eventfd(struct fd_parms *p, int lfd, const struct cr_fdset *set)
 	return do_dump_gen_file(p, lfd, set);
 }
 
+static int dump_eventpoll(struct fd_parms *p, int lfd, const struct cr_fdset *set)
+{
+	p->id = MAKE_FD_GENID(p->stat.st_dev, p->stat.st_ino, p->pos);
+	p->type	= FDINFO_EVENTPOLL;
+	return do_dump_gen_file(p, lfd, set);
+}
+
 static int dump_one_file(pid_t pid, int fd, int lfd, char fd_flags,
 		       const struct cr_fdset *cr_fdset)
 {
@@ -489,6 +500,8 @@ static int dump_one_file(pid_t pid, int fd, int lfd, char fd_flags,
 	if (is_anon_inode(&statfs)) {
 		if (is_eventfd_link(lfd))
 			return dump_eventfd(&p, lfd, cr_fdset);
+		else if (is_eventpoll_link(lfd))
+			return dump_eventpoll(&p, lfd, cr_fdset);
 	}
 
 	if (S_ISREG(p.stat.st_mode) ||

cr-restore.c

@@ -36,6 +36,7 @@
 #include "pipes.h"
 #include "sk-inet.h"
 #include "eventfd.h"
+#include "eventpoll.h"
 #include "proc_parse.h"
 #include "restorer-blob.h"
 #include "crtools.h"
@@ -166,6 +167,9 @@ static int prepare_shared(void)
 	if (collect_eventfd())
 		return -1;
 
+	if (collect_eventpoll())
+		return -1;
+
 	list_for_each_entry(pi, &tasks, list) {
 		ret = prepare_shmem_pid(pi->pid);
 		if (ret < 0)

cr-show.c

@@ -63,6 +63,7 @@ static char *fdtype2s(u8 type)
 		[FDINFO_PIPE] = "pipe",
 		[FDINFO_UNIXSK] = "usk",
 		[FDINFO_EVENTFD] = "efd",
+		[FDINFO_EVENTPOLL] = "epl",
 	};
 
 	if (type > FDINFO_UND && type < FD_INFO_MAX)

crtools.c

@@ -25,6 +25,7 @@
 #include "files.h"
 #include "sk-inet.h"
 #include "eventfd.h"
+#include "eventpoll.h"
 
 struct cr_options opts;
 
@@ -71,6 +72,20 @@ struct cr_fd_desc_tmpl fdset_template[CR_FD_MAX] = {
 		.show	= show_eventfds,
 	},
 
+	/* eventpoll */
+	[CR_FD_EVENTPOLL] = {
+		.fmt	= FMT_FNAME_EVENTPOLL,
+		.magic	= EVENTPOLL_MAGIC,
+		.show	= show_eventpoll,
+	},
+
+	/* eventpoll target file descriptors */
+	[CR_FD_EVENTPOLL_TFD] = {
+		.fmt	= FMT_FNAME_EVENTPOLL_TFD,
+		.magic	= EVENTPOLL_TFD_MAGIC,
+		.show	= show_eventpoll_tfd,
+	},
+
 	/* core data, such as regs and vmas and such */
 	[CR_FD_CORE] = {
 		.fmt	= FMT_FNAME_CORE,

eventpoll.c

+#include <unistd.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <errno.h>
+#include <fcntl.h>
+#include <stdlib.h>
+#include <string.h>
+#include <limits.h>
+#include <sys/stat.h>
+#include <sys/statfs.h>
+#include <sys/types.h>
+#include <sys/ioctl.h>
+#include <sys/epoll.h>
+
+#include "compiler.h"
+#include "types.h"
+#include "eventpoll.h"
+
+#include "crtools.h"
+#include "image.h"
+#include "util.h"
+#include "log.h"
+
+struct eventpoll_file_info {
+	struct eventpoll_file_entry	efe;
+	struct file_desc		d;
+};
+
+struct eventpoll_tfd_file_info {
+	struct eventpoll_tfd_entry	tdefe;
+	struct list_head		list;
+};
+
+static LIST_HEAD(eventpoll_tfds);
+
+/* Checks if file desciptor @lfd is eventfd */
+int is_eventpoll_link(int lfd)
+{
+	char link[PATH_MAX], path[32];
+	ssize_t ret;
+
+	snprintf(path, sizeof(path), "/proc/self/fd/%d", lfd);
+	ret = readlink(path, link, sizeof(link));
+	if (ret < 0) {
+		pr_perror("Can't read link of fd %d\n", lfd);
+		return 0;
+	}
+	link[ret] = 0;
+	if (!strcmp(link, "anon_inode:[eventpoll]"))
+		return 1;
+
+	return 0;
+}
+
+static void pr_info_eventpoll_tfd(char *action, struct eventpoll_tfd_entry *e)
+{
+	pr_info("%seventpoll-tfd: id %#08x tfd %#08x events %#08x data %#016lx\n",
+		action, e->id, e->tfd, e->events, e->data);
+}
+
+static void pr_info_eventpoll(char *action, struct eventpoll_file_entry *e)
+{
+	pr_info("%seventpoll: id %#08x flags %#04x\n", action, e->id, e->flags);
+}
+
+void show_eventpoll_tfd(int fd, struct cr_options *o)
+{
+	struct eventpoll_tfd_entry e;
+
+	pr_img_head(CR_FD_EVENTPOLL_TFD);
+
+	while (1) {
+		int ret;
+
+		ret = read_img_eof(fd, &e);
+		if (ret <= 0)
+			goto out;
+		pr_msg("id: %#08x tfd %#08x events %#08x data %#016lx\n",
+		       e.id, e.tfd, e.events, e.data);
+	}
+
+out:
+	pr_img_tail(CR_FD_EVENTPOLL_TFD);
+}
+
+void show_eventpoll(int fd, struct cr_options *o)
+{
+	struct eventpoll_file_entry e;
+
+	pr_img_head(CR_FD_EVENTPOLL);
+
+	while (1) {
+		int ret;
+
+		ret = read_img_eof(fd, &e);
+		if (ret <= 0)
+			goto out;
+		pr_msg("id: %#08x flags %#04x ",
+		       e.id, e.flags);
+		show_fown_cont(&e.fown);
+		pr_msg("\n");
+	}
+
+out:
+	pr_img_tail(CR_FD_EVENTPOLL);
+}
+
+int dump_one_eventpoll(int lfd, u32 id, const struct fd_parms *p)
+{
+	int image_fd = fdset_fd(glob_fdset, CR_FD_EVENTPOLL);
+	int image_tfd = fdset_fd(glob_fdset, CR_FD_EVENTPOLL_TFD);
+	struct eventpoll_file_entry e;
+	struct eventpoll_tfd_entry efd;
+	char buf[PAGE_SIZE], *tok;
+	int ret, fdinfo;
+
+	snprintf(buf, sizeof(buf), "/proc/self/fdinfo/%d", lfd);
+	fdinfo = open(buf, O_RDONLY);
+	if (fdinfo < 0) {
+		pr_perror("Can't open %d (%d)", p->fd, lfd);
+		return -1;
+	}
+
+	ret = read(fdinfo, buf, sizeof(buf));
+	close(fdinfo);
+	if (ret <= 0) {
+		pr_perror("Reading eventpoll from %d (%d) failed", p->fd, lfd);
+		return -1;
+	}
+
+	e.id	= id;
+	e.flags	= p->flags;
+	e.fown	= p->fown;
+
+	pr_info_eventpoll("Dumping ", &e);
+	if (write_img(image_fd, &e))
+		return -1;
+
+	tok = strstr(buf, "tfd:");
+	if (!tok)
+		goto parsing_err;
+
+	tok = strtok(tok, "\n");
+	while (tok) {
+		efd.id = id;
+		if (sscanf(tok, "tfd: %8d events: %8x data: %16lx",
+			&efd.tfd, &efd.events, &efd.data) != 3)
+			goto parsing_err;
+		tok = strtok(NULL, "\n");
+
+		pr_info_eventpoll_tfd("Dumping: ", &efd);
+		if (write_img(image_tfd, &efd))
+			return -1;
+	}
+
+	return 0;
+
+parsing_err:
+	pr_err("Parsing error %d (%d)", p->fd, lfd);
+	return -1;
+}
+
+static int eventpoll_open(struct file_desc *d)
+{
+	struct eventpoll_tfd_file_info *td_info;
+	struct eventpoll_file_info *info;
+	int tmp, ret;
+
+	info = container_of(d, struct eventpoll_file_info, d);
+
+	tmp = epoll_create(1);
+	if (tmp < 0) {
+		pr_perror("Can't create epoll %#08x",
+			  info->efe.id);
+		return -1;
+	}
+
+	if (rst_file_params(tmp, &info->efe.fown, info->efe.flags)) {
+		pr_perror("Can't restore file params on epoll %#08x",
+			  info->efe.id);
+		goto err_close;
+	}
+
+	list_for_each_entry(td_info, &eventpoll_tfds, list) {
+		struct epoll_event event;
+
+		if (td_info->tdefe.id != info->efe.id)
+			continue;
+
+		event.events	= td_info->tdefe.events;
+		event.data.u64	= td_info->tdefe.data;
+		ret = epoll_ctl(tmp, EPOLL_CTL_ADD, td_info->tdefe.tfd, &event);
+		if (ret) {
+			pr_perror("Can't add event on %#08x", info->efe.id);
+			goto err_close;
+		}
+	}
+
+	return tmp;
+
+err_close:
+	close(tmp);
+	return -1;
+}
+
+static struct file_desc_ops desc_ops = {
+	.open = eventpoll_open,
+};
+
+int collect_eventpoll(void)
+{
+	int image_fd;
+	int ret = -1;
+
+	image_fd = open_image_ro(CR_FD_EVENTPOLL_TFD);
+	if (image_fd < 0)
+		return -1;
+
+	while (1) {
+		struct eventpoll_tfd_file_info *info;
+
+		info = xmalloc(sizeof(*info));
+		if (!info)
+			goto err;
+
+		ret = read_img_eof(image_fd, &info->tdefe);
+		if (ret < 0)
+			goto err;
+		else if (!ret)
+			break;
+
+		INIT_LIST_HEAD(&info->list);
+
+		list_add(&info->list, &eventpoll_tfds);
+		pr_info_eventpoll_tfd("Collected ", &info->tdefe);
+	}
+
+	close_safe(&image_fd);
+
+	image_fd = open_image_ro(CR_FD_EVENTPOLL);
+	if (image_fd < 0)
+		return -1;
+
+	while (1) {
+		struct eventpoll_file_info *info;
+
+		info = xmalloc(sizeof(*info));
+		if (!info)
+			goto err;
+
+		ret = read_img_eof(image_fd, &info->efe);
+		if (ret < 0)
+			goto err;
+		else if (!ret)
+			break;
+
+		pr_info_eventpoll("Collected ", &info->efe);
+		file_desc_add(&info->d, FDINFO_EVENTPOLL, info->efe.id, &desc_ops);
+	}
+
+err:
+	close_safe(&image_fd);
+	return ret;
+}

include/crtools.h

@@ -56,6 +56,8 @@ enum {
 	CR_FD_PIPES_DATA,
 	CR_FD_REMAP_FPATH,
 	CR_FD_EVENTFD,
+	CR_FD_EVENTPOLL,
+	CR_FD_EVENTPOLL_TFD,
 	_CR_FD_GLOB_TO,
 
 	CR_FD_MAX
@@ -122,6 +124,8 @@ extern struct cr_fd_desc_tmpl fdset_template[CR_FD_MAX];
 #define FMT_FNAME_UNIXSK	"unixsk.img"
 #define FMT_FNAME_INETSK	"inetsk.img"
 #define FMT_FNAME_EVENTFD	"eventfd.img"
+#define FMT_FNAME_EVENTPOLL	"eventpoll.img"
+#define FMT_FNAME_EVENTPOLL_TFD	"eventpoll-tfd.img"
 #define FMT_FNAME_ITIMERS	"itimers-%d.img"
 #define FMT_FNAME_CREDS		"creds-%d.img"
 #define FMT_FNAME_UTSNS		"utsns-%d.img"

include/eventpoll.h

+#ifndef EVENTPOLL_H__
+#define EVENTPOLL_H__
+
+#include <sys/types.h>
+#include <unistd.h>
+
+#include "compiler.h"
+#include "types.h"
+#include "files.h"
+#include "crtools.h"
+
+extern int is_eventpoll_link(int lfd);
+extern int dump_one_eventpoll(int lfd, u32 id, const struct fd_parms *p);
+extern int collect_eventpoll(void);
+extern void show_eventpoll(int fd, struct cr_options *o);
+extern void show_eventpoll_tfd(int fd, struct cr_options *o);
+
+#endif /* EVENTPOLL_H__ */

include/image.h

@@ -34,6 +34,8 @@
 #define GHOST_FILE_MAGIC	0x52583605 /* Oryol */
 #define TCP_STREAM_MAGIC	0x51465506 /* Orenburg */
 #define EVENTFD_MAGIC		0x44523722 /* Anapa */
+#define EVENTPOLL_MAGIC		0x45023858 /* Krasnodar */
+#define EVENTPOLL_TFD_MAGIC	0x44433746 /* Novorossiysk */
 
 #define PIPEFS_MAGIC	0x50495045
 
@@ -44,6 +46,7 @@ enum fd_types {
 	FDINFO_INETSK,
 	FDINFO_UNIXSK,
 	FDINFO_EVENTFD,
+	FDINFO_EVENTPOLL,
 
 	FD_INFO_MAX
 };
@@ -93,6 +96,19 @@ struct eventfd_file_entry {
 	u64	counter;
 } __packed;
 
+struct eventpoll_tfd_entry {
+	u32	id;
+	u32	tfd;
+	u32	events;
+	u64	data;
+} __packed;
+
+struct eventpoll_file_entry {
+	u32	id;
+	u32	flags;
+	fown_t	fown;
+} __packed;
+
 struct fdinfo_entry {
 	u32	fd;
 	u8	type;